diff --git a/apps/database-new/.env.local.example b/apps/database-new/.env.local.example
index 70bad2e5610..eaad2b0e239 100644
--- a/apps/database-new/.env.local.example
+++ b/apps/database-new/.env.local.example
@@ -3,3 +3,5 @@
 NEXT_PUBLIC_SUPABASE_URL=your-project-url
 NEXT_PUBLIC_SUPABASE_ANON_KEY=your-anon-key
 OPENAI_API_KEY=your-openai-api-key
+OPENAI_BASE_URL=your-openai-compatible-base-url
+OPENAI_MODEL=your-language-model
diff --git a/apps/database-new/app/[thread_id]/[message_id]/getAssistantMessage.ts b/apps/database-new/app/[thread_id]/[message_id]/getAssistantMessage.ts
index 5d5de4d165a..b5acb6e01f9 100644
--- a/apps/database-new/app/[thread_id]/[message_id]/getAssistantMessage.ts
+++ b/apps/database-new/app/[thread_id]/[message_id]/getAssistantMessage.ts
@@ -1,17 +1,12 @@
 import { stripIndent } from 'common-tags'
-import { cookies } from 'next/headers'
 import OpenAI from 'openai'
 
 // import is weird, what's up with this?
 import { createClient } from '@/lib/supabase/server'
+import { createOpenAiClient } from 'ai-commands'
 import { ContextLengthError } from '../../../../../packages/ai-commands/src/errors'
 import { getMessages } from './getMessages'
 
-// Create an OpenAI API client (that's edge friendly!)
-const openai = new OpenAI({
-  apiKey: process.env.OPENAI_API_KEY,
-})
-
 // Set the runtime to edge for best performance
 export const runtime = 'edge'
 
@@ -26,6 +21,12 @@ export type AiAssistantMessage = {
  * it.
  */
 export const getAssistantResponse = async (threadId: string, messageId: string) => {
+  const { openai, model, error: openAiError } = createOpenAiClient()
+
+  if (openAiError) {
+    throw openAiError
+  }
+
   const supabase = createClient()
 
   const { data: message, error } = await supabase
@@ -91,7 +92,7 @@ export const getAssistantResponse = async (threadId: string, messageId: string)
 
     try {
       const response = await openai.chat.completions.create({
-        model: 'gpt-3.5-turbo-16k-0613',
+        model,
         messages: initMessages,
         max_tokens: 1024,
         temperature: 0,
diff --git a/apps/database-new/package.json b/apps/database-new/package.json
index d7ed811868e..5b37491c5ae 100644
--- a/apps/database-new/package.json
+++ b/apps/database-new/package.json
@@ -16,6 +16,7 @@
     "@supabase/ssr": "^0.1.0",
     "@tanstack/react-query": "^5.13.4",
     "ai": "^2.2.31",
+    "ai-commands": "*",
     "common": "*",
     "config": "*",
     "dayjs": "^1.11.10",
@@ -42,4 +43,4 @@
     "postcss": "^8.4.31",
     "typescript": "^5.4.3"
   }
-}
+}
\ No newline at end of file
diff --git a/apps/studio/.env b/apps/studio/.env
index 4de7d26d76e..8dd533d0f4a 100644
--- a/apps/studio/.env
+++ b/apps/studio/.env
@@ -6,7 +6,6 @@ DEFAULT_PROJECT_NAME=Default Project
 DASHBOARD_USERNAME=supabase
 DASHBOARD_PASSWORD=this_password_is_insecure_and_should_be_updated
 
-
 SUPABASE_URL=http://localhost:8000
 SUPABASE_PUBLIC_URL=http://localhost:8000
 SUPABASE_ANON_KEY=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyAgCiAgICAicm9sZSI6ICJhbm9uIiwKICAgICJpc3MiOiAic3VwYWJhc2UtZGVtbyIsCiAgICAiaWF0IjogMTY0MTc2OTIwMCwKICAgICJleHAiOiAxNzk5NTM1NjAwCn0.dc_X5iR_VP_qT0zsiyj_I_OZ2T9FtRU2BBNWN8Bu4GE
@@ -24,4 +23,11 @@ NEXT_PUBLIC_HCAPTCHA_SITE_KEY=10000000-ffff-ffff-ffff-000000000001
 NEXT_PUBLIC_SUPABASE_ANON_KEY=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJpc3MiOiJzdXBhYmFzZSIsInJlZiI6InhndWloeHV6cWlid3hqbmlteGV2Iiwicm9sZSI6ImFub24iLCJpYXQiOjE2NzUwOTQ4MzUsImV4cCI6MTk5MDY3MDgzNX0.0PMlOxtKL4O9GGZuAP_Xl4f-Tut1qOnW4bNEmAtoB8w
 NEXT_PUBLIC_SUPABASE_URL=https://xguihxuzqibwxjnimxev.supabase.co
 
-DOCKER_SOCKET_LOCATION=/var/run/docker.sock
\ No newline at end of file
+OPENAI_API_KEY="your-openai-api-key"
+OPENAI_BASE_URL="your-openai-compatible-base-url"
+OPENAI_MODEL="your-language-model"
+OPENAI_EMBEDDING_API_KEY="your-openai-api-key"
+OPENAI_EMBEDDING_BASE_URL="your-openai-compatible-base-url"
+OPENAI_EMBEDDING_MODEL="your-embedding-model"
+
+DOCKER_SOCKET_LOCATION=/var/run/docker.sock
diff --git a/apps/studio/pages/api/ai/sql/debug.ts b/apps/studio/pages/api/ai/sql/debug.ts
index 965bf49f142..06fb59dd62d 100644
--- a/apps/studio/pages/api/ai/sql/debug.ts
+++ b/apps/studio/pages/api/ai/sql/debug.ts
@@ -1,18 +1,9 @@
 import { ContextLengthError, EmptySqlError, debugSql } from 'ai-commands'
+import { createOpenAiClient } from 'ai-commands/src/openai'
 import apiWrapper from 'lib/api/apiWrapper'
 import { NextApiRequest, NextApiResponse } from 'next'
-import { OpenAI } from 'openai'
-
-const openAiKey = process.env.OPENAI_KEY
-const openai = new OpenAI({ apiKey: openAiKey })
 
 async function handler(req: NextApiRequest, res: NextApiResponse) {
-  if (!openAiKey) {
-    return res.status(500).json({
-      error: 'No OPENAI_KEY set. Create this environment variable to use AI features.',
-    })
-  }
-
   const { method } = req
 
   switch (method) {
@@ -25,12 +16,18 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
 }
 
 export async function handlePost(req: NextApiRequest, res: NextApiResponse) {
+  const { openai, model, error } = createOpenAiClient()
+
+  if (error) {
+    return res.status(500).json({ error: error.message })
+  }
+
   const {
     body: { errorMessage, sql, entityDefinitions },
   } = req
 
   try {
-    const result = await debugSql(openai, errorMessage, sql, entityDefinitions)
+    const result = await debugSql(openai, model, errorMessage, sql, entityDefinitions)
     return res.json(result)
   } catch (error) {
     if (error instanceof Error) {
diff --git a/apps/studio/pages/api/ai/sql/edit.ts b/apps/studio/pages/api/ai/sql/edit.ts
index 3e564c19185..8acc8df2019 100644
--- a/apps/studio/pages/api/ai/sql/edit.ts
+++ b/apps/studio/pages/api/ai/sql/edit.ts
@@ -1,18 +1,9 @@
 import { ContextLengthError, EmptySqlError, editSql } from 'ai-commands'
+import { createOpenAiClient } from 'ai-commands/src/openai'
 import apiWrapper from 'lib/api/apiWrapper'
 import { NextApiRequest, NextApiResponse } from 'next'
-import { OpenAI } from 'openai'
-
-const openAiKey = process.env.OPENAI_KEY
-const openai = new OpenAI({ apiKey: openAiKey })
 
 async function handler(req: NextApiRequest, res: NextApiResponse) {
-  if (!openAiKey) {
-    return res.status(500).json({
-      error: 'No OPENAI_KEY set. Create this environment variable to use AI features.',
-    })
-  }
-
   const { method } = req
 
   switch (method) {
@@ -25,12 +16,18 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
 }
 
 export async function handlePost(req: NextApiRequest, res: NextApiResponse) {
+  const { openai, model, error } = createOpenAiClient()
+
+  if (error) {
+    return res.status(500).json({ error: error.message })
+  }
+
   const {
     body: { prompt, sql, entityDefinitions },
   } = req
 
   try {
-    const result = await editSql(openai, prompt, sql, entityDefinitions)
+    const result = await editSql(openai, model, prompt, sql, entityDefinitions)
     return res.json(result)
   } catch (error) {
     if (error instanceof Error) {
diff --git a/apps/studio/pages/api/ai/sql/generate-v2.ts b/apps/studio/pages/api/ai/sql/generate-v2.ts
index 144f01a93bd..27b4823623e 100644
--- a/apps/studio/pages/api/ai/sql/generate-v2.ts
+++ b/apps/studio/pages/api/ai/sql/generate-v2.ts
@@ -1,25 +1,11 @@
 import { StreamingTextResponse } from 'ai'
 import { generateV2 } from 'ai-commands/edge'
+import { createOpenAiClient } from 'ai-commands/src/openai'
 import { NextRequest } from 'next/server'
-import OpenAI from 'openai'
 
 export const runtime = 'edge'
 
-const openAiKey = process.env.OPENAI_KEY
-
 export default async function handler(req: NextRequest) {
-  if (!openAiKey) {
-    return new Response(
-      JSON.stringify({
-        error: 'No OPENAI_KEY set. Create this environment variable to use AI features.',
-      }),
-      {
-        status: 500,
-        headers: { 'Content-Type': 'application/json' },
-      }
-    )
-  }
-
   const { method } = req
 
   switch (method) {
@@ -37,7 +23,19 @@ export default async function handler(req: NextRequest) {
 }
 
 async function handlePost(request: NextRequest) {
-  const openai = new OpenAI({ apiKey: openAiKey })
+  const { openai, model, error } = createOpenAiClient()
+
+  if (error) {
+    return new Response(
+      JSON.stringify({
+        error: error.message,
+      }),
+      {
+        status: 500,
+        headers: { 'Content-Type': 'application/json' },
+      }
+    )
+  }
 
   const body = await (request.json() as Promise<{
     messages: { content: string; role: 'user' | 'assistant' }[]
@@ -48,7 +46,7 @@ async function handlePost(request: NextRequest) {
   const { messages, existingSql, entityDefinitions } = body
 
   try {
-    const stream = await generateV2(openai, messages, existingSql, entityDefinitions)
+    const stream = await generateV2(openai, model, messages, existingSql, entityDefinitions)
     return new StreamingTextResponse(stream)
   } catch (error) {
     if (error instanceof Error) {
diff --git a/apps/studio/pages/api/ai/sql/generate.ts b/apps/studio/pages/api/ai/sql/generate.ts
index 1dc73696ac3..86848aed248 100644
--- a/apps/studio/pages/api/ai/sql/generate.ts
+++ b/apps/studio/pages/api/ai/sql/generate.ts
@@ -1,18 +1,9 @@
 import { ContextLengthError, EmptySqlError, generateSql } from 'ai-commands'
+import { createOpenAiClient } from 'ai-commands/src/openai'
 import apiWrapper from 'lib/api/apiWrapper'
 import { NextApiRequest, NextApiResponse } from 'next'
-import { OpenAI } from 'openai'
-
-const openAiKey = process.env.OPENAI_KEY
-const openai = new OpenAI({ apiKey: openAiKey })
 
 async function handler(req: NextApiRequest, res: NextApiResponse) {
-  if (!openAiKey) {
-    return res.status(500).json({
-      error: 'No OPENAI_KEY set. Create this environment variable to use AI features.',
-    })
-  }
-
   const { method } = req
 
   switch (method) {
@@ -25,12 +16,18 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
 }
 
 export async function handlePost(req: NextApiRequest, res: NextApiResponse) {
+  const { openai, model, error } = createOpenAiClient()
+
+  if (error) {
+    return res.status(500).json({ error: error.message })
+  }
+
   const {
     body: { prompt, entityDefinitions },
   } = req
 
   try {
-    const result = await generateSql(openai, prompt, entityDefinitions)
+    const result = await generateSql(openai, model, prompt, entityDefinitions)
     return res.json(result)
   } catch (error) {
     if (error instanceof Error) {
diff --git a/apps/studio/pages/api/ai/sql/suggest.ts b/apps/studio/pages/api/ai/sql/suggest.ts
index 3f60da2f8ff..757ed6bf7f3 100644
--- a/apps/studio/pages/api/ai/sql/suggest.ts
+++ b/apps/studio/pages/api/ai/sql/suggest.ts
@@ -1,25 +1,11 @@
 import { StreamingTextResponse } from 'ai'
 import { chatRlsPolicy } from 'ai-commands/edge'
+import { createOpenAiClient } from 'ai-commands/src/openai'
 import { NextRequest } from 'next/server'
-import OpenAI from 'openai'
 
 export const runtime = 'edge'
 
-const openAiKey = process.env.OPENAI_KEY
-
 export default async function handler(req: NextRequest) {
-  if (!openAiKey) {
-    return new Response(
-      JSON.stringify({
-        error: 'No OPENAI_KEY set. Create this environment variable to use AI features.',
-      }),
-      {
-        status: 500,
-        headers: { 'Content-Type': 'application/json' },
-      }
-    )
-  }
-
   const { method } = req
 
   switch (method) {
@@ -37,7 +23,19 @@ export default async function handler(req: NextRequest) {
 }
 
 async function handlePost(request: NextRequest) {
-  const openai = new OpenAI({ apiKey: openAiKey })
+  const { openai, model, error } = createOpenAiClient()
+
+  if (error) {
+    return new Response(
+      JSON.stringify({
+        error: error.message,
+      }),
+      {
+        status: 500,
+        headers: { 'Content-Type': 'application/json' },
+      }
+    )
+  }
 
   const body = await (request.json() as Promise<{
     messages: { content: string; role: 'user' | 'assistant' }[]
@@ -48,7 +46,7 @@ async function handlePost(request: NextRequest) {
   const { messages, entityDefinitions, policyDefinition } = body
 
   try {
-    const stream = await chatRlsPolicy(openai, messages, entityDefinitions, policyDefinition)
+    const stream = await chatRlsPolicy(openai, model, messages, entityDefinitions, policyDefinition)
     return new StreamingTextResponse(stream)
   } catch (error) {
     console.error(error)
diff --git a/apps/studio/pages/api/ai/sql/title.ts b/apps/studio/pages/api/ai/sql/title.ts
index c5d1477faf9..bd153e35ca4 100644
--- a/apps/studio/pages/api/ai/sql/title.ts
+++ b/apps/studio/pages/api/ai/sql/title.ts
@@ -1,18 +1,9 @@
 import { ContextLengthError, titleSql } from 'ai-commands'
+import { createOpenAiClient } from 'ai-commands/src/openai'
 import apiWrapper from 'lib/api/apiWrapper'
 import { NextApiRequest, NextApiResponse } from 'next'
-import { OpenAI } from 'openai'
-
-const openAiKey = process.env.OPENAI_KEY
-const openai = new OpenAI({ apiKey: openAiKey })
 
 async function handler(req: NextApiRequest, res: NextApiResponse) {
-  if (!openAiKey) {
-    return res.status(500).json({
-      error: 'No OPENAI_KEY set. Create this environment variable to use AI features.',
-    })
-  }
-
   const { method } = req
 
   switch (method) {
@@ -25,12 +16,18 @@ async function handler(req: NextApiRequest, res: NextApiResponse) {
 }
 
 export async function handlePost(req: NextApiRequest, res: NextApiResponse) {
+  const { openai, model, error } = createOpenAiClient()
+
+  if (error) {
+    return res.status(500).json({ error: error.message })
+  }
+
   const {
     body: { sql },
   } = req
 
   try {
-    const result = await titleSql(openai, sql)
+    const result = await titleSql(openai, model, sql)
     return res.json(result)
   } catch (error) {
     if (error instanceof Error) {
diff --git a/package-lock.json b/package-lock.json
index d249a3026b2..69dca12ca82 100644
--- a/package-lock.json
+++ b/package-lock.json
@@ -39,6 +39,7 @@
         "@supabase/ssr": "^0.1.0",
         "@tanstack/react-query": "^5.13.4",
         "ai": "^2.2.31",
+        "ai-commands": "*",
         "common": "*",
         "config": "*",
         "dayjs": "^1.11.10",
@@ -6983,6 +6984,14 @@
         "react-hook-form": "^7.0.0"
       }
     },
+    "node_modules/@huggingface/jinja": {
+      "version": "0.2.2",
+      "resolved": "https://registry.npmjs.org/@huggingface/jinja/-/jinja-0.2.2.tgz",
+      "integrity": "sha512-/KPde26khDUIPkTGU82jdtTW9UAuvUTumCAbFs/7giR0SxsvZC4hru51PBvpijH6BVkHcROcvZM/lpy5h1jRRA==",
+      "engines": {
+        "node": ">=18"
+      }
+    },
     "node_modules/@humanwhocodes/config-array": {
       "version": "0.11.14",
       "resolved": "https://registry.npmjs.org/@humanwhocodes/config-array/-/config-array-0.11.14.tgz",
@@ -8646,6 +8655,60 @@
         "url": "https://opencollective.com/popperjs"
       }
     },
+    "node_modules/@protobufjs/aspromise": {
+      "version": "1.1.2",
+      "resolved": "https://registry.npmjs.org/@protobufjs/aspromise/-/aspromise-1.1.2.tgz",
+      "integrity": "sha512-j+gKExEuLmKwvz3OgROXtrJ2UG2x8Ch2YZUxahh+s1F2HZ+wAceUNLkvy6zKCPVRkU++ZWQrdxsUeQXmcg4uoQ=="
+    },
+    "node_modules/@protobufjs/base64": {
+      "version": "1.1.2",
+      "resolved": "https://registry.npmjs.org/@protobufjs/base64/-/base64-1.1.2.tgz",
+      "integrity": "sha512-AZkcAA5vnN/v4PDqKyMR5lx7hZttPDgClv83E//FMNhR2TMcLUhfRUBHCmSl0oi9zMgDDqRUJkSxO3wm85+XLg=="
+    },
+    "node_modules/@protobufjs/codegen": {
+      "version": "2.0.4",
+      "resolved": "https://registry.npmjs.org/@protobufjs/codegen/-/codegen-2.0.4.tgz",
+      "integrity": "sha512-YyFaikqM5sH0ziFZCN3xDC7zeGaB/d0IUb9CATugHWbd1FRFwWwt4ld4OYMPWu5a3Xe01mGAULCdqhMlPl29Jg=="
+    },
+    "node_modules/@protobufjs/eventemitter": {
+      "version": "1.1.0",
+      "resolved": "https://registry.npmjs.org/@protobufjs/eventemitter/-/eventemitter-1.1.0.tgz",
+      "integrity": "sha512-j9ednRT81vYJ9OfVuXG6ERSTdEL1xVsNgqpkxMsbIabzSo3goCjDIveeGv5d03om39ML71RdmrGNjG5SReBP/Q=="
+    },
+    "node_modules/@protobufjs/fetch": {
+      "version": "1.1.0",
+      "resolved": "https://registry.npmjs.org/@protobufjs/fetch/-/fetch-1.1.0.tgz",
+      "integrity": "sha512-lljVXpqXebpsijW71PZaCYeIcE5on1w5DlQy5WH6GLbFryLUrBD4932W/E2BSpfRJWseIL4v/KPgBFxDOIdKpQ==",
+      "dependencies": {
+        "@protobufjs/aspromise": "^1.1.1",
+        "@protobufjs/inquire": "^1.1.0"
+      }
+    },
+    "node_modules/@protobufjs/float": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/@protobufjs/float/-/float-1.0.2.tgz",
+      "integrity": "sha512-Ddb+kVXlXst9d+R9PfTIxh1EdNkgoRe5tOX6t01f1lYWOvJnSPDBlG241QLzcyPdoNTsblLUdujGSE4RzrTZGQ=="
+    },
+    "node_modules/@protobufjs/inquire": {
+      "version": "1.1.0",
+      "resolved": "https://registry.npmjs.org/@protobufjs/inquire/-/inquire-1.1.0.tgz",
+      "integrity": "sha512-kdSefcPdruJiFMVSbn801t4vFK7KB/5gd2fYvrxhuJYg8ILrmn9SKSX2tZdV6V+ksulWqS7aXjBcRXl3wHoD9Q=="
+    },
+    "node_modules/@protobufjs/path": {
+      "version": "1.1.2",
+      "resolved": "https://registry.npmjs.org/@protobufjs/path/-/path-1.1.2.tgz",
+      "integrity": "sha512-6JOcJ5Tm08dOHAbdR3GrvP+yUUfkjG5ePsHYczMFLq3ZmMkAD98cDgcT2iA1lJ9NVwFd4tH/iSSoe44YWkltEA=="
+    },
+    "node_modules/@protobufjs/pool": {
+      "version": "1.1.0",
+      "resolved": "https://registry.npmjs.org/@protobufjs/pool/-/pool-1.1.0.tgz",
+      "integrity": "sha512-0kELaGSIDBKvcgS4zkjz1PeddatrjYcmMWOlAuAPwAeccUrPHdUqo/J6LiymHHEiJT5NrF1UVwxY14f+fy4WQw=="
+    },
+    "node_modules/@protobufjs/utf8": {
+      "version": "1.1.0",
+      "resolved": "https://registry.npmjs.org/@protobufjs/utf8/-/utf8-1.1.0.tgz",
+      "integrity": "sha512-Vvn3zZrhQZkkBE8LSuW3em98c0FwgO4nxzv6OdSxPKJIEKY2bGbHn+mhGIPerzI4twdxaP8/0+06HBpwf345Lw=="
+    },
     "node_modules/@radix-ui/colors": {
       "version": "0.1.9",
       "license": "MIT"
@@ -17012,6 +17075,11 @@
       "integrity": "sha512-OvlIYQK9tNneDlS0VN54LLd5uiPCBOp7gS5Z0f1mjoJYBrtStzgmJBxONW3U6OZqdtNzZPmn9BS/7WI7BFFcFQ==",
       "dev": true
     },
+    "node_modules/@types/long": {
+      "version": "4.0.2",
+      "resolved": "https://registry.npmjs.org/@types/long/-/long-4.0.2.tgz",
+      "integrity": "sha512-MqTGEo5bj5t157U6fA/BiDynNkn0YknVdh48CMPkTSpFTVmvao5UQmm7uEF6xBEo7qIMAlY/JSleYaE6VOdpaA=="
+    },
     "node_modules/@types/markdown-table": {
       "version": "3.0.0",
       "dev": true,
@@ -18041,6 +18109,19 @@
         "@xtuc/long": "4.2.2"
       }
     },
+    "node_modules/@xenova/transformers": {
+      "version": "2.17.1",
+      "resolved": "https://registry.npmjs.org/@xenova/transformers/-/transformers-2.17.1.tgz",
+      "integrity": "sha512-zo702tQAFZXhzeD2GCYUNUqeqkoueOdiSbQWa4s0q7ZE4z8WBIwIsMMPGobpgdqjQ2u0Qulo08wuqVEUrBXjkQ==",
+      "dependencies": {
+        "@huggingface/jinja": "^0.2.2",
+        "onnxruntime-web": "1.14.0",
+        "sharp": "^0.32.0"
+      },
+      "optionalDependencies": {
+        "onnxruntime-node": "1.14.0"
+      }
+    },
     "node_modules/@xobotyi/scrollbar-width": {
       "version": "1.9.5",
       "license": "MIT"
@@ -18898,8 +18979,7 @@
     "node_modules/b4a": {
       "version": "1.6.6",
       "resolved": "https://registry.npmjs.org/b4a/-/b4a-1.6.6.tgz",
-      "integrity": "sha512-5Tk1HLk6b6ctmjIkAcU/Ujv/1WqiDl0F0JdRCR80VsOcUlHcu7pWeWRlOqQLHfDEsVx9YH/aif5AG4ehoCtTmg==",
-      "dev": true
+      "integrity": "sha512-5Tk1HLk6b6ctmjIkAcU/Ujv/1WqiDl0F0JdRCR80VsOcUlHcu7pWeWRlOqQLHfDEsVx9YH/aif5AG4ehoCtTmg=="
     },
     "node_modules/babel-core": {
       "version": "7.0.0-bridge.0",
@@ -19345,14 +19425,12 @@
       "version": "2.2.0",
       "resolved": "https://registry.npmjs.org/bare-events/-/bare-events-2.2.0.tgz",
       "integrity": "sha512-Yyyqff4PIFfSuthCZqLlPISTWHmnQxoPuAvkmgzsJEmG3CesdIv6Xweayl0JkCZJSB2yYIdJyEz97tpxNhgjbg==",
-      "dev": true,
       "optional": true
     },
     "node_modules/bare-fs": {
       "version": "2.2.0",
       "resolved": "https://registry.npmjs.org/bare-fs/-/bare-fs-2.2.0.tgz",
       "integrity": "sha512-+VhW202E9eTVGkX7p+TNXtZC4RTzj9JfJW7PtfIbZ7mIQ/QT9uOafQTx7lx2n9ERmWsXvLHF4hStAFn4gl2mQw==",
-      "dev": true,
       "optional": true,
       "dependencies": {
         "bare-events": "^2.0.0",
@@ -19365,14 +19443,12 @@
       "version": "2.2.0",
       "resolved": "https://registry.npmjs.org/bare-os/-/bare-os-2.2.0.tgz",
       "integrity": "sha512-hD0rOPfYWOMpVirTACt4/nK8mC55La12K5fY1ij8HAdfQakD62M+H4o4tpfKzVGLgRDTuk3vjA4GqGXXCeFbag==",
-      "dev": true,
       "optional": true
     },
     "node_modules/bare-path": {
       "version": "2.1.0",
       "resolved": "https://registry.npmjs.org/bare-path/-/bare-path-2.1.0.tgz",
       "integrity": "sha512-DIIg7ts8bdRKwJRJrUMy/PICEaQZaPGZ26lsSx9MJSwIhSrcdHn7/C8W+XmnG/rKi6BaRcz+JO00CjZteybDtw==",
-      "dev": true,
       "optional": true,
       "dependencies": {
         "bare-os": "^2.1.0"
@@ -22207,7 +22283,6 @@
       "version": "6.0.0",
       "resolved": "https://registry.npmjs.org/decompress-response/-/decompress-response-6.0.0.tgz",
       "integrity": "sha512-aW35yZM6Bb/4oJlZncMH2LCoZtJXTRxES17vE3hoRiowU2kWHaJKFkSBDnDR+cm9J+9QhXmREyIfv0pji9ejCQ==",
-      "dev": true,
       "dependencies": {
         "mimic-response": "^3.1.0"
       },
@@ -22280,7 +22355,6 @@
       "version": "0.6.0",
       "resolved": "https://registry.npmjs.org/deep-extend/-/deep-extend-0.6.0.tgz",
       "integrity": "sha512-LOHxIOaPYdHlJRtCQfDIVZtfw/ufM8+rVj649RIHzcm/vGwQRXFt6OPqIFWsm2XEMrNIEtWR64sY1LEKD2vAOA==",
-      "dev": true,
       "engines": {
         "node": ">=4.0.0"
       }
@@ -24336,7 +24410,6 @@
       "version": "2.0.3",
       "resolved": "https://registry.npmjs.org/expand-template/-/expand-template-2.0.3.tgz",
       "integrity": "sha512-XYfuKMvj4O35f/pOXLObndIRvyQ+/+6AhODh+OKWj9S9498pHHn/IMszH+gt0fBCRWMNfk1ZSp5x3AifmnI2vg==",
-      "dev": true,
       "engines": {
         "node": ">=6"
       }
@@ -24533,8 +24606,7 @@
     "node_modules/fast-fifo": {
       "version": "1.3.2",
       "resolved": "https://registry.npmjs.org/fast-fifo/-/fast-fifo-1.3.2.tgz",
-      "integrity": "sha512-/d9sfos4yxzpwkDkuN7k2SqFKtYNmCTzgfEpz82x34IM9/zc8KGxQoXg1liNC/izpRM/MBdt44Nmx41ZWqk+FQ==",
-      "dev": true
+      "integrity": "sha512-/d9sfos4yxzpwkDkuN7k2SqFKtYNmCTzgfEpz82x34IM9/zc8KGxQoXg1liNC/izpRM/MBdt44Nmx41ZWqk+FQ=="
     },
     "node_modules/fast-glob": {
       "version": "3.3.2",
@@ -24889,6 +24961,11 @@
         "node": ">=12.0.0"
       }
     },
+    "node_modules/flatbuffers": {
+      "version": "1.12.0",
+      "resolved": "https://registry.npmjs.org/flatbuffers/-/flatbuffers-1.12.0.tgz",
+      "integrity": "sha512-c7CZADjRcl6j0PlvFy0ZqXQ67qSEZfrVPynmnL+2zPc+NtMvrF8Y0QceMo7QqnSPc7+uWjUIAbvCQ5WIKlMVdQ=="
+    },
     "node_modules/flatted": {
       "version": "3.2.9",
       "license": "ISC"
@@ -25502,8 +25579,7 @@
     "node_modules/github-from-package": {
       "version": "0.0.0",
       "resolved": "https://registry.npmjs.org/github-from-package/-/github-from-package-0.0.0.tgz",
-      "integrity": "sha512-SyHy3T1v2NUXn29OsWdxmK6RwHD+vkj3v8en8AOBZ1wBQ/hCAQ5bAQTD02kW4W9tUp/3Qh6J8r9EvntiyCmOOw==",
-      "dev": true
+      "integrity": "sha512-SyHy3T1v2NUXn29OsWdxmK6RwHD+vkj3v8en8AOBZ1wBQ/hCAQ5bAQTD02kW4W9tUp/3Qh6J8r9EvntiyCmOOw=="
     },
     "node_modules/github-slugger": {
       "version": "2.0.0",
@@ -25683,6 +25759,11 @@
         "node": ">=6.0"
       }
     },
+    "node_modules/guid-typescript": {
+      "version": "1.0.9",
+      "resolved": "https://registry.npmjs.org/guid-typescript/-/guid-typescript-1.0.9.tgz",
+      "integrity": "sha512-Y8T4vYhEfwJOTbouREvG+3XDsjr8E3kIr7uf+JZ0BYloFsttiHU0WfvANVsR7TxNUJa/WpCnw/Ino/p+DeBhBQ=="
+    },
     "node_modules/gulp-header": {
       "version": "1.8.12",
       "license": "MIT",
@@ -27100,8 +27181,7 @@
     "node_modules/ini": {
       "version": "1.3.8",
       "resolved": "https://registry.npmjs.org/ini/-/ini-1.3.8.tgz",
-      "integrity": "sha512-JV/yugV2uzW5iMRSiZAyDtQd+nxtUnjeLt0acNdw98kKLrvuRVyB80tsREOE7yvGVgalhZ6RNXCmEHkUKBKxew==",
-      "dev": true
+      "integrity": "sha512-JV/yugV2uzW5iMRSiZAyDtQd+nxtUnjeLt0acNdw98kKLrvuRVyB80tsREOE7yvGVgalhZ6RNXCmEHkUKBKxew=="
     },
     "node_modules/ini-simple-parser": {
       "version": "1.0.0",
@@ -30404,6 +30484,11 @@
         "node": ">=8"
       }
     },
+    "node_modules/long": {
+      "version": "4.0.0",
+      "resolved": "https://registry.npmjs.org/long/-/long-4.0.0.tgz",
+      "integrity": "sha512-XsP+KhQif4bjX1kbuSiySJFNAehNxgLb6hPRGJ9QsUr8ajHkuXGdrHmFUTUUXhDwVX2R5bY4JNZEwbUiMhV+MA=="
+    },
     "node_modules/longest-streak": {
       "version": "3.1.0",
       "license": "MIT",
@@ -34588,7 +34673,6 @@
       "version": "3.1.0",
       "resolved": "https://registry.npmjs.org/mimic-response/-/mimic-response-3.1.0.tgz",
       "integrity": "sha512-z0yWI+4FDrrweS8Zmt4Ej5HdJmky15+L2e6Wgn3+iK5fWzb6T3fhNFq2+MeTRb064c6Wr4N/wv0DzQTjNzHNGQ==",
-      "dev": true,
       "engines": {
         "node": ">=10"
       },
@@ -35015,8 +35099,7 @@
     "node_modules/napi-build-utils": {
       "version": "1.0.2",
       "resolved": "https://registry.npmjs.org/napi-build-utils/-/napi-build-utils-1.0.2.tgz",
-      "integrity": "sha512-ONmRUqK7zj7DWX0D9ADe03wbwOBZxNAfF20PlGfCWQcD3+/MakShIHrMqx9YwPTfxDdF1zLeL+RGZiR9kGMLdg==",
-      "dev": true
+      "integrity": "sha512-ONmRUqK7zj7DWX0D9ADe03wbwOBZxNAfF20PlGfCWQcD3+/MakShIHrMqx9YwPTfxDdF1zLeL+RGZiR9kGMLdg=="
     },
     "node_modules/natural-compare": {
       "version": "1.4.0",
@@ -35197,7 +35280,6 @@
       "version": "3.56.0",
       "resolved": "https://registry.npmjs.org/node-abi/-/node-abi-3.56.0.tgz",
       "integrity": "sha512-fZjdhDOeRcaS+rcpve7XuwHBmktS1nS1gzgghwKUQQ8nTy2FdSDr6ZT8k6YhvlJeHmmQMYiT/IH9hfco5zeW2Q==",
-      "dev": true,
       "dependencies": {
         "semver": "^7.3.5"
       },
@@ -35209,7 +35291,6 @@
       "version": "6.0.0",
       "resolved": "https://registry.npmjs.org/lru-cache/-/lru-cache-6.0.0.tgz",
       "integrity": "sha512-Jo6dJ04CmSjuznwJSS3pUeWmd/H0ffTlkXXgwZi+eq1UCmqQwCh+eLsYOYCwY991i2Fah4h1BEMCx4qThGbsiA==",
-      "dev": true,
       "dependencies": {
         "yallist": "^4.0.0"
       },
@@ -35221,7 +35302,6 @@
       "version": "7.6.0",
       "resolved": "https://registry.npmjs.org/semver/-/semver-7.6.0.tgz",
       "integrity": "sha512-EnwXhrlwXMk9gKu5/flx5sv/an57AkRplG3hTK68W7FRDN+k+OWBj65M7719OkA82XLBxrcX0KSHj+X5COhOVg==",
-      "dev": true,
       "dependencies": {
         "lru-cache": "^6.0.0"
       },
@@ -35235,8 +35315,7 @@
     "node_modules/node-abi/node_modules/yallist": {
       "version": "4.0.0",
       "resolved": "https://registry.npmjs.org/yallist/-/yallist-4.0.0.tgz",
-      "integrity": "sha512-3wdGidZyq5PB084XLES5TpOSRA3wjXAlIWMhum2kRcv/41Sn2emQ0dycQW4uZXLejwKvg6EsvbdlVL+FYEct7A==",
-      "dev": true
+      "integrity": "sha512-3wdGidZyq5PB084XLES5TpOSRA3wjXAlIWMhum2kRcv/41Sn2emQ0dycQW4uZXLejwKvg6EsvbdlVL+FYEct7A=="
     },
     "node_modules/node-abort-controller": {
       "version": "3.1.1",
@@ -36364,6 +36443,46 @@
         "url": "https://github.com/sponsors/sindresorhus"
       }
     },
+    "node_modules/onnx-proto": {
+      "version": "4.0.4",
+      "resolved": "https://registry.npmjs.org/onnx-proto/-/onnx-proto-4.0.4.tgz",
+      "integrity": "sha512-aldMOB3HRoo6q/phyB6QRQxSt895HNNw82BNyZ2CMh4bjeKv7g/c+VpAFtJuEMVfYLMbRx61hbuqnKceLeDcDA==",
+      "dependencies": {
+        "protobufjs": "^6.8.8"
+      }
+    },
+    "node_modules/onnxruntime-common": {
+      "version": "1.14.0",
+      "resolved": "https://registry.npmjs.org/onnxruntime-common/-/onnxruntime-common-1.14.0.tgz",
+      "integrity": "sha512-3LJpegM2iMNRX2wUmtYfeX/ytfOzNwAWKSq1HbRrKc9+uqG/FsEA0bbKZl1btQeZaXhC26l44NWpNUeXPII7Ew=="
+    },
+    "node_modules/onnxruntime-node": {
+      "version": "1.14.0",
+      "resolved": "https://registry.npmjs.org/onnxruntime-node/-/onnxruntime-node-1.14.0.tgz",
+      "integrity": "sha512-5ba7TWomIV/9b6NH/1x/8QEeowsb+jBEvFzU6z0T4mNsFwdPqXeFUM7uxC6QeSRkEbWu3qEB0VMjrvzN/0S9+w==",
+      "optional": true,
+      "os": [
+        "win32",
+        "darwin",
+        "linux"
+      ],
+      "dependencies": {
+        "onnxruntime-common": "~1.14.0"
+      }
+    },
+    "node_modules/onnxruntime-web": {
+      "version": "1.14.0",
+      "resolved": "https://registry.npmjs.org/onnxruntime-web/-/onnxruntime-web-1.14.0.tgz",
+      "integrity": "sha512-Kcqf43UMfW8mCydVGcX9OMXI2VN17c0p6XvR7IPSZzBf/6lteBzXHvcEVWDPmCKuGombl997HgLqj91F11DzXw==",
+      "dependencies": {
+        "flatbuffers": "^1.12.0",
+        "guid-typescript": "^1.0.9",
+        "long": "^4.0.0",
+        "onnx-proto": "^4.0.4",
+        "onnxruntime-common": "~1.14.0",
+        "platform": "^1.3.6"
+      }
+    },
     "node_modules/open": {
       "version": "8.4.2",
       "resolved": "https://registry.npmjs.org/open/-/open-8.4.2.tgz",
@@ -37123,6 +37242,11 @@
         "pathe": "^1.1.0"
       }
     },
+    "node_modules/platform": {
+      "version": "1.3.6",
+      "resolved": "https://registry.npmjs.org/platform/-/platform-1.3.6.tgz",
+      "integrity": "sha512-fnWVljUchTro6RiCFvCXBbNhJc2NijN7oIQxbwsyL0buWJPG85v81ehlHI9fXrJsMNgTofEoWIQeClKpgxFLrg=="
+    },
     "node_modules/playwright": {
       "version": "1.41.2",
       "resolved": "https://registry.npmjs.org/playwright/-/playwright-1.41.2.tgz",
@@ -37560,7 +37684,6 @@
       "version": "7.1.1",
       "resolved": "https://registry.npmjs.org/prebuild-install/-/prebuild-install-7.1.1.tgz",
       "integrity": "sha512-jAXscXWMcCK8GgCoHOfIr0ODh5ai8mj63L2nWrjuAgXE6tDyYGnx4/8o/rCgU+B4JSyZBKbeZqzhtwtC3ovxjw==",
-      "dev": true,
       "dependencies": {
         "detect-libc": "^2.0.0",
         "expand-template": "^2.0.3",
@@ -37852,6 +37975,31 @@
         "url": "https://github.com/sponsors/wooorm"
       }
     },
+    "node_modules/protobufjs": {
+      "version": "6.11.4",
+      "resolved": "https://registry.npmjs.org/protobufjs/-/protobufjs-6.11.4.tgz",
+      "integrity": "sha512-5kQWPaJHi1WoCpjTGszzQ32PG2F4+wRY6BmAT4Vfw56Q2FZ4YZzK20xUYQH4YkfehY1e6QSICrJquM6xXZNcrw==",
+      "hasInstallScript": true,
+      "dependencies": {
+        "@protobufjs/aspromise": "^1.1.2",
+        "@protobufjs/base64": "^1.1.2",
+        "@protobufjs/codegen": "^2.0.4",
+        "@protobufjs/eventemitter": "^1.1.0",
+        "@protobufjs/fetch": "^1.1.0",
+        "@protobufjs/float": "^1.0.2",
+        "@protobufjs/inquire": "^1.1.0",
+        "@protobufjs/path": "^1.1.2",
+        "@protobufjs/pool": "^1.1.0",
+        "@protobufjs/utf8": "^1.1.0",
+        "@types/long": "^4.0.1",
+        "@types/node": ">=13.7.0",
+        "long": "^4.0.0"
+      },
+      "bin": {
+        "pbjs": "bin/pbjs",
+        "pbts": "bin/pbts"
+      }
+    },
     "node_modules/proxy-addr": {
       "version": "2.0.7",
       "dev": true,
@@ -38132,8 +38280,7 @@
     "node_modules/queue-tick": {
       "version": "1.0.1",
       "resolved": "https://registry.npmjs.org/queue-tick/-/queue-tick-1.0.1.tgz",
-      "integrity": "sha512-kJt5qhMxoszgU/62PLP1CJytzd2NKetjSRnyuj31fDd3Rlcz3fzlFdFLD1SItunPwyqEOkca6GbV612BWfaBag==",
-      "dev": true
+      "integrity": "sha512-kJt5qhMxoszgU/62PLP1CJytzd2NKetjSRnyuj31fDd3Rlcz3fzlFdFLD1SItunPwyqEOkca6GbV612BWfaBag=="
     },
     "node_modules/raf-schd": {
       "version": "4.0.3",
@@ -38246,7 +38393,6 @@
       "version": "1.2.8",
       "resolved": "https://registry.npmjs.org/rc/-/rc-1.2.8.tgz",
       "integrity": "sha512-y3bGgqKj3QBdxLbLkomlohkvsA8gdAiUQlSBJnBhfn+BPxg4bc62d8TcBW15wavDfgexCgccckhcZvywyQYPOw==",
-      "dev": true,
       "dependencies": {
         "deep-extend": "^0.6.0",
         "ini": "~1.3.0",
@@ -38261,7 +38407,6 @@
       "version": "2.0.1",
       "resolved": "https://registry.npmjs.org/strip-json-comments/-/strip-json-comments-2.0.1.tgz",
       "integrity": "sha512-4gB8na07fecVVkOI6Rs4e7T6NOTki5EmL7TUduTs6bu3EdnSycntVJ4re8kgZA+wx9IueI2Y11bfbgwtzuE0KQ==",
-      "dev": true,
       "engines": {
         "node": ">=0.10.0"
       }
@@ -41423,7 +41568,6 @@
       "version": "0.32.6",
       "resolved": "https://registry.npmjs.org/sharp/-/sharp-0.32.6.tgz",
       "integrity": "sha512-KyLTWwgcR9Oe4d9HwCwNM2l7+J0dUQwn/yf7S0EnTtb0eVS4RxO0eUSvxPtzT4F3SY+C4K6fqdv/DO27sJ/v/w==",
-      "dev": true,
       "hasInstallScript": true,
       "dependencies": {
         "color": "^4.2.3",
@@ -41446,7 +41590,6 @@
       "version": "6.0.0",
       "resolved": "https://registry.npmjs.org/lru-cache/-/lru-cache-6.0.0.tgz",
       "integrity": "sha512-Jo6dJ04CmSjuznwJSS3pUeWmd/H0ffTlkXXgwZi+eq1UCmqQwCh+eLsYOYCwY991i2Fah4h1BEMCx4qThGbsiA==",
-      "dev": true,
       "dependencies": {
         "yallist": "^4.0.0"
       },
@@ -41457,14 +41600,12 @@
     "node_modules/sharp/node_modules/node-addon-api": {
       "version": "6.1.0",
       "resolved": "https://registry.npmjs.org/node-addon-api/-/node-addon-api-6.1.0.tgz",
-      "integrity": "sha512-+eawOlIgy680F0kBzPUNFhMZGtJ1YmqM6l4+Crf4IkImjYrO/mqPwRMh352g23uIaQKFItcQ64I7KMaJxHgAVA==",
-      "dev": true
+      "integrity": "sha512-+eawOlIgy680F0kBzPUNFhMZGtJ1YmqM6l4+Crf4IkImjYrO/mqPwRMh352g23uIaQKFItcQ64I7KMaJxHgAVA=="
     },
     "node_modules/sharp/node_modules/semver": {
       "version": "7.6.0",
       "resolved": "https://registry.npmjs.org/semver/-/semver-7.6.0.tgz",
       "integrity": "sha512-EnwXhrlwXMk9gKu5/flx5sv/an57AkRplG3hTK68W7FRDN+k+OWBj65M7719OkA82XLBxrcX0KSHj+X5COhOVg==",
-      "dev": true,
       "dependencies": {
         "lru-cache": "^6.0.0"
       },
@@ -41479,7 +41620,6 @@
       "version": "3.0.5",
       "resolved": "https://registry.npmjs.org/tar-fs/-/tar-fs-3.0.5.tgz",
       "integrity": "sha512-JOgGAmZyMgbqpLwct7ZV8VzkEB6pxXFBVErLtb+XCOqzc6w1xiWKI9GVd6bwk68EX7eJ4DWmfXVmq8K2ziZTGg==",
-      "dev": true,
       "dependencies": {
         "pump": "^3.0.0",
         "tar-stream": "^3.1.5"
@@ -41493,7 +41633,6 @@
       "version": "3.1.7",
       "resolved": "https://registry.npmjs.org/tar-stream/-/tar-stream-3.1.7.tgz",
       "integrity": "sha512-qJj60CXt7IU1Ffyc3NJMjh6EkuCFej46zUqJ4J7pqYlThyd9bO0XBTmcOIhSzZJVWfsLks0+nle/j538YAW9RQ==",
-      "dev": true,
       "dependencies": {
         "b4a": "^1.6.4",
         "fast-fifo": "^1.2.0",
@@ -41503,8 +41642,7 @@
     "node_modules/sharp/node_modules/yallist": {
       "version": "4.0.0",
       "resolved": "https://registry.npmjs.org/yallist/-/yallist-4.0.0.tgz",
-      "integrity": "sha512-3wdGidZyq5PB084XLES5TpOSRA3wjXAlIWMhum2kRcv/41Sn2emQ0dycQW4uZXLejwKvg6EsvbdlVL+FYEct7A==",
-      "dev": true
+      "integrity": "sha512-3wdGidZyq5PB084XLES5TpOSRA3wjXAlIWMhum2kRcv/41Sn2emQ0dycQW4uZXLejwKvg6EsvbdlVL+FYEct7A=="
     },
     "node_modules/shebang-command": {
       "version": "2.0.0",
@@ -41619,7 +41757,6 @@
       "version": "1.0.1",
       "resolved": "https://registry.npmjs.org/simple-concat/-/simple-concat-1.0.1.tgz",
       "integrity": "sha512-cSFtAPtRhljv69IK0hTVZQ+OfE9nePi/rtJmw5UjHeVyVroEqJXP1sFztKUy1qU+xvz3u/sfYJLa947b7nAN2Q==",
-      "dev": true,
       "funding": [
         {
           "type": "github",
@@ -41639,7 +41776,6 @@
       "version": "4.0.1",
       "resolved": "https://registry.npmjs.org/simple-get/-/simple-get-4.0.1.tgz",
       "integrity": "sha512-brv7p5WgH0jmQJr1ZDDfKDOSeWWg+OVypG99A/5vYGPqJ6pxiaHLy8nxtFjBA7oMa01ebA9gfh1uMCFqOuXxvA==",
-      "dev": true,
       "funding": [
         {
           "type": "github",
@@ -42301,7 +42437,6 @@
       "version": "2.16.1",
       "resolved": "https://registry.npmjs.org/streamx/-/streamx-2.16.1.tgz",
       "integrity": "sha512-m9QYj6WygWyWa3H1YY69amr4nVgy61xfjys7xO7kviL5rfIEc2naf+ewFiOA+aEJD7y0JO3h2GoiUv4TDwEGzQ==",
-      "dev": true,
       "dependencies": {
         "fast-fifo": "^1.1.0",
         "queue-tick": "^1.0.1"
@@ -44445,7 +44580,6 @@
       "version": "0.6.0",
       "resolved": "https://registry.npmjs.org/tunnel-agent/-/tunnel-agent-0.6.0.tgz",
       "integrity": "sha512-McnNiV1l8RYeY8tBgEpuodCC1mLUdbSN+CYBL7kJsJNInOP8UjDDEwdk6Mw60vdLLrr5NHKZhMAOSrR2NZuQ+w==",
-      "dev": true,
       "dependencies": {
         "safe-buffer": "^5.0.1"
       },
@@ -47343,6 +47477,7 @@
         "@gregnr/libpg-query": "^13.4.0-dev.12",
         "@serafin/schema-builder": "^0.18.5",
         "@supabase/supabase-js": "*",
+        "@xenova/transformers": "^2.17.1",
         "ai": "^2.2.29",
         "common-tags": "^1.8.2",
         "config": "*",
diff --git a/packages/ai-commands/index.ts b/packages/ai-commands/index.ts
index dc525721fcc..5aee37c5b17 100644
--- a/packages/ai-commands/index.ts
+++ b/packages/ai-commands/index.ts
@@ -1,2 +1,4 @@
 export * from './src/errors'
+export * from './src/openai'
 export * from './src/sql'
+export * from './src/tokenizer'
diff --git a/packages/ai-commands/jest.config.js b/packages/ai-commands/jest.config.js
index b970fa8370a..8b619eb8803 100644
--- a/packages/ai-commands/jest.config.js
+++ b/packages/ai-commands/jest.config.js
@@ -10,6 +10,6 @@ module.exports = {
   setupFilesAfterEnv: ['./test/extensions.ts'],
   testTimeout: 15000,
   transformIgnorePatterns: [
-    'node_modules/(?!(mdast-.*|micromark|micromark-.*|unist-.*|decode-named-character-reference|character-entities)/)',
+    'node_modules/(?!(mdast-.*|micromark|micromark-.*|unist-.*|decode-named-character-reference|character-entities|@xenova/transformers)/)',
   ],
 }
diff --git a/packages/ai-commands/package.json b/packages/ai-commands/package.json
index fe02d5959bc..1524bd0d865 100644
--- a/packages/ai-commands/package.json
+++ b/packages/ai-commands/package.json
@@ -12,6 +12,7 @@
     "@gregnr/libpg-query": "^13.4.0-dev.12",
     "@serafin/schema-builder": "^0.18.5",
     "@supabase/supabase-js": "*",
+    "@xenova/transformers": "^2.17.1",
     "ai": "^2.2.29",
     "common-tags": "^1.8.2",
     "config": "*",
diff --git a/packages/ai-commands/src/__snapshots__/sql.test.ts.snap b/packages/ai-commands/src/__snapshots__/sql.test.ts.snap
index ed775186dac..48eed7c9f8a 100644
--- a/packages/ai-commands/src/__snapshots__/sql.test.ts.snap
+++ b/packages/ai-commands/src/__snapshots__/sql.test.ts.snap
@@ -32,9 +32,9 @@ exports[`edit add length constraint 1`] = `
 exports[`generate single table with specified columns 1`] = `
 "create table employees (
   id bigint primary key generated always as identity,
-  name text,
-  email text,
-  position text
+  name text not null,
+  email text unique not null,
+  position text not null
 );"
 `;
 
@@ -46,3 +46,25 @@ exports[`rls chat select policy using table definition 1`] = `
   "usingExpression": "user_id = auth.uid ()",
 }
 `;
+
+exports[`sql chat single table with specified columns 1`] = `
+"create table employees (
+  id bigint primary key generated always as identity,
+  name text not null,
+  email text unique not null,
+  position text not null,
+  created_at timestamp with time zone default current_timestamp,
+  updated_at timestamp with time zone default current_timestamp
+);"
+`;
+
+exports[`sql chat table referencing users creates fk to auth.users 1`] = `
+"create table todos (
+  id bigint primary key generated always as identity,
+  title text not null,
+  description text,
+  created_at timestamp with time zone default current_timestamp,
+  author_id uuid references auth.users (id),
+  completed boolean default false
+);"
+`;
diff --git a/packages/ai-commands/src/openai.ts b/packages/ai-commands/src/openai.ts
new file mode 100644
index 00000000000..443a80184ea
--- /dev/null
+++ b/packages/ai-commands/src/openai.ts
@@ -0,0 +1,129 @@
+import OpenAI from 'openai'
+
+export const supportedModels = [
+  'llama3',
+  'llama3-70b-8192',
+  'llama3-8b-8192',
+  'gpt-4-turbo',
+  'gpt-4-turbo-preview',
+  'gpt-4-turbo-2024-04-09',
+  'gpt-4-0125-preview',
+  'gpt-4-1106-preview',
+  'gpt-4',
+  'gpt-4-0613',
+  'gpt-4-0314',
+  'gpt-3.5-turbo',
+  'gpt-3.5-turbo-0125',
+  'gpt-3.5-turbo-1106',
+]
+
+export const supportedEmbeddingModels = ['text-embedding-ada-002']
+
+/**
+ * Instantiates OpenAI client while checking and validating
+ * environment variables.
+ *
+ * Allows you to use your own OpenAI-compatible provider.
+ *
+ * Uses the following env vars:
+ * - OPENAI_API_KEY (backwards compatible with previous OPENAI_KEY)
+ * - OPENAI_BASE_URL
+ * - OPENAI_MODEL
+ */
+export function createOpenAiClient() {
+  const apiKey = process.env.OPENAI_API_KEY ?? process.env.OPENAI_KEY
+  const baseURL = process.env.OPENAI_BASE_URL
+  const model = process.env.OPENAI_MODEL
+
+  if (!apiKey) {
+    return {
+      error: new Error(
+        'No OPENAI_API_KEY set. Create this environment variable to use AI features.'
+      ),
+    }
+  }
+
+  if (!baseURL) {
+    return {
+      error: new Error(
+        'No OPENAI_BASE_URL set. Create this environment variable to use AI features.'
+      ),
+    }
+  }
+
+  if (!model) {
+    return {
+      error: new Error('No OPENAI_MODEL set. Create this environment variable to use AI features.'),
+    }
+  }
+
+  if (!supportedModels.includes(model)) {
+    return {
+      error: new Error(
+        `The following models are supported for OPENAI_MODEL: ${JSON.stringify(supportedModels)}`
+      ),
+    }
+  }
+
+  const openai = new OpenAI({ apiKey, baseURL })
+
+  return { openai, model }
+}
+
+/**
+ * Instantiates OpenAI client for embedding models while checking and validating
+ * environment variables.
+ *
+ * Allows you to use separate OpenAI-compatible providers for embedding models vs. language models.
+ *
+ * Uses the following env vars:
+ * - OPENAI_EMBEDDING_API_KEY
+ * - OPENAI_EMBEDDING_BASE_URL
+ * - OPENAI_EMBEDDING_MODEL
+ *
+ * Falls back to these default env vars if the above vars don't exist:
+ * - OPENAI_API_KEY
+ * - OPENAI_BASE_URL
+ */
+export function createOpenAiEmbeddingClient() {
+  const apiKey =
+    process.env.OPENAI_EMBEDDING_API_KEY ?? process.env.OPENAI_API_KEY ?? process.env.OPENAI_KEY
+  const baseURL = process.env.OPENAI_EMBEDDING_BASE_URL ?? process.env.OPENAI_BASE_URL
+  const model = process.env.OPENAI_EMBEDDING_MODEL
+
+  if (!apiKey) {
+    return {
+      error: new Error(
+        'No OPENAI_EMBEDDING_API_KEY set. Create this environment variable to use AI embedding features.'
+      ),
+    }
+  }
+
+  if (!baseURL) {
+    return {
+      error: new Error(
+        'No OPENAI_EMBEDDING_BASE_URL set. Create this environment variable to use AI embedding features.'
+      ),
+    }
+  }
+
+  if (!model) {
+    return {
+      error: new Error(
+        'No OPENAI_EMBEDDING_MODEL set. Create this environment variable to use AI embedding features.'
+      ),
+    }
+  }
+
+  if (!supportedEmbeddingModels.includes(model)) {
+    return {
+      error: new Error(
+        `The following models are supported for OPENAI_EMBEDDING_MODEL: ${JSON.stringify(supportedEmbeddingModels)}`
+      ),
+    }
+  }
+
+  const openai = new OpenAI({ apiKey, baseURL })
+
+  return { openai, model }
+}
diff --git a/packages/ai-commands/src/sql.edge.ts b/packages/ai-commands/src/sql.edge.ts
index fb5164e9bc2..f34d27ae7c6 100644
--- a/packages/ai-commands/src/sql.edge.ts
+++ b/packages/ai-commands/src/sql.edge.ts
@@ -3,13 +3,8 @@ import { OpenAIStream } from 'ai'
 import { codeBlock, oneLine, stripIndent } from 'common-tags'
 import OpenAI from 'openai'
 
-import { ApplicationError, ContextLengthError, UserError } from './errors'
-import { getChatRequestTokenCount, getMaxTokenCount, tokenizer } from './tokenizer'
-
-interface Message {
-  role: 'user' | 'assistant'
-  content: string
-}
+import { ApplicationError, ContextLengthError } from './errors'
+import { Message, capMessages, countTokens } from './tokenizer'
 
 /**
  * Responds to a conversation about building an RLS policy.
@@ -18,6 +13,7 @@ interface Message {
  */
 export async function chatRlsPolicy(
   openai: OpenAI,
+  model: string,
   messages: Message[],
   entityDefinitions?: string[],
   policyDefinition?: string
@@ -26,7 +22,7 @@ export async function chatRlsPolicy(
     {
       role: 'system',
       content: stripIndent`
-        You're an Postgres expert in writing row level security policies. Your purpose is to 
+        You're a Postgres expert in writing row level security policies. Your purpose is to 
         generate a policy with the constraints given by the user. You will be provided a schema 
         on which the policy should be applied.
 
@@ -35,18 +31,18 @@ export async function chatRlsPolicy(
         - Always use double apostrophe in SQL strings (eg. 'Night''s watch')
         - You can use only CREATE POLICY or ALTER POLICY queries, no other queries are allowed.
         - You can add short explanations to your messages.
-        - The result should be a valid markdown. The SQL code should be wrapped in \`\`\`.
+        - The result should be a valid markdown. The SQL code should be wrapped in \`\`\` (including sql language tag).
         - Always use "auth.uid()" instead of "current_user".
         - You can't use "USING" expression on INSERT policies.
         - Only use "WITH CHECK" expression on INSERT or UPDATE policies.
-        - The policy name should be short text explaining the policy, enclosed in double quotes.
+        - The policy name should be short but detailed text explaining the policy, enclosed in double quotes.
         - Always put explanations as separate text. Never use inline SQL comments. 
         - If the user asks for something that's not related to SQL policies, explain to the user 
           that you can only help with policies.
         
         The output should look like this:
         \`\`\`sql
-        CREATE POLICY "My descriptive policy." ON users FOR INSERT USING (user_name = current_user) WITH (true);
+        CREATE POLICY "My descriptive policy." ON books FOR INSERT USING (author_id = auth.uid()) WITH (true);
         \`\`\`
       `,
     },
@@ -77,7 +73,7 @@ export async function chatRlsPolicy(
 
   try {
     const response = await openai.chat.completions.create({
-      model: 'gpt-3.5-turbo-0125',
+      model,
       messages: initMessages,
       max_tokens: 1024,
       temperature: 0,
@@ -101,24 +97,37 @@ export async function chatRlsPolicy(
  */
 export async function generateV2(
   openai: OpenAI,
+  model: string,
   messages: Message[],
   existingSql?: string,
   entityDefinitions?: string[]
 ): Promise<ReadableStream<Uint8Array>> {
-  const initMessages: OpenAI.Chat.Completions.ChatCompletionMessageParam[] = [
+  const initMessages: Message[] = [
     {
       role: 'system',
       content: stripIndent`
+      You are a Supabase Postgres SQL generator. The user will have a conversation with you and your job is to generate SQL
+      in the Postgres dialect. Be friendly and relatable, complementing the user when appropriate, but don't go overboard.
+
       The generated SQL (must be valid SQL).
       - For primary keys, always use "id bigint primary key generated always as identity" (not serial)
       - Prefer creating foreign key references in the create statement
+      - When creating new tables, always add foreign key references inline (ie. "my_column my_type references ..." on a single line)
+      - Include all necessary tables and create them in the correct order (eg. book references author, so create author before book)
       - Prefer 'text' over 'varchar'
       - Prefer 'timestamp with time zone' over 'date'
       - Use vector(384) data type for any embedding/vector related query
       - Always use double apostrophe in SQL strings (eg. 'Night''s watch')
       - Always use semicolons
+      - There is a built-in auth.users table with a uuid id column (but it's immutable)
+        - Prefer referencing auth.users directly via FK instead of creating a new users table
+        - Only create your own users table if you need to store additional data about the user such as avatar
+        - If you do create your own users table, prefer to call it "profiles" and add a FK to auth.users. You must also create a trigger function
+          that automatically creates the profile when there is a new auth.user
+      - Sprinkle in detailed explanations throughout your answer, digging into "why" each snippet is necessary
+        - Explanations should live outside of code blocks
       - Output as markdown
-      - Always include code snippets if available
+      - Always include code snippets if available (including sql language tag, ie. \`\`\`sql\n...\n\`\`\`)
       `,
     },
   ]
@@ -148,7 +157,7 @@ export async function generateV2(
 
   try {
     const response = await openai.chat.completions.create({
-      model: 'gpt-3.5-turbo-0125',
+      model,
       messages: initMessages,
       max_tokens: 1024,
       temperature: 0,
@@ -167,6 +176,9 @@ export async function generateV2(
 
 export async function clippy(
   openai: OpenAI,
+  openaiEmbedding: OpenAI,
+  model: string,
+  embeddingModel: string,
   supabaseClient: SupabaseClient<any, 'public', any>,
   messages: Message[]
 ) {
@@ -188,25 +200,9 @@ export async function clippy(
     throw new Error("No message with role 'user'")
   }
 
-  // Moderate the content to comply with OpenAI T&C
-  const moderationResponses = await Promise.all(
-    contextMessages.map((message) => openai.moderations.create({ input: message.content }))
-  )
-
-  for (const moderationResponse of moderationResponses) {
-    const [results] = moderationResponse.results
-
-    if (results.flagged) {
-      throw new UserError('Flagged content', {
-        flagged: true,
-        categories: results.categories,
-      })
-    }
-  }
-
-  const embeddingResponse = await openai.embeddings
+  const embeddingResponse = await openaiEmbedding.embeddings
     .create({
-      model: 'text-embedding-ada-002',
+      model: embeddingModel,
       input: userMessage.content.replaceAll('\n', ' '),
     })
     .catch((error: any) => {
@@ -235,8 +231,7 @@ export async function clippy(
   for (let i = 0; i < pageSections.length; i++) {
     const pageSection = pageSections[i]
     const content = pageSection.content
-    const encoded = tokenizer.encode(content)
-    tokenCount += encoded.length
+    tokenCount += await countTokens(content, model)
 
     if (tokenCount >= 1500) {
       break
@@ -245,7 +240,7 @@ export async function clippy(
     contextText += `${content.trim()}\n---\n`
   }
 
-  const initMessages: OpenAI.Chat.Completions.ChatCompletionMessageParam[] = [
+  const initMessages: Message[] = [
     {
       role: 'system',
       content: codeBlock`
@@ -307,20 +302,19 @@ export async function clippy(
     },
   ]
 
-  const model = 'gpt-3.5-turbo-0301'
-  const maxCompletionTokenCount = 1024
+  const maxCompletionTokens = 1024
 
-  const completionMessages: OpenAI.Chat.Completions.ChatCompletionMessageParam[] = capMessages(
+  const completionMessages: Message[] = await capMessages(
     initMessages,
     contextMessages,
-    maxCompletionTokenCount,
-    model
+    model,
+    maxCompletionTokens
   )
 
   const completionOptions = {
     model,
     messages: completionMessages,
-    max_tokens: 1024,
+    max_tokens: maxCompletionTokens,
     temperature: 0,
     stream: true,
   }
@@ -342,32 +336,3 @@ export async function clippy(
 
   return response
 }
-
-/**
- * Remove context messages until the entire request fits
- * the max total token count for that model.
- *
- * Accounts for both message and completion token counts.
- */
-function capMessages(
-  initMessages: OpenAI.Chat.Completions.ChatCompletionMessageParam[],
-  contextMessages: OpenAI.Chat.Completions.ChatCompletionMessageParam[],
-  maxCompletionTokenCount: number,
-  model: string
-) {
-  const maxTotalTokenCount = getMaxTokenCount(model)
-  const cappedContextMessages = [...contextMessages]
-  let tokenCount =
-    getChatRequestTokenCount([...initMessages, ...cappedContextMessages], model) +
-    maxCompletionTokenCount
-
-  // Remove earlier context messages until we fit
-  while (tokenCount >= maxTotalTokenCount) {
-    cappedContextMessages.shift()
-    tokenCount =
-      getChatRequestTokenCount([...initMessages, ...cappedContextMessages], model) +
-      maxCompletionTokenCount
-  }
-
-  return [...initMessages, ...cappedContextMessages]
-}
diff --git a/packages/ai-commands/src/sql.test.ts b/packages/ai-commands/src/sql.test.ts
index 730832d51ce..1306a0e2dc0 100644
--- a/packages/ai-commands/src/sql.test.ts
+++ b/packages/ai-commands/src/sql.test.ts
@@ -1,17 +1,23 @@
 import { describe, expect, test } from '@jest/globals'
 import { codeBlock } from 'common-tags'
-import OpenAI from 'openai'
 import { collectStream, extractMarkdownSql, formatSql, getPolicyInfo } from '../test/util'
+import { createOpenAiClient } from './openai'
 import { debugSql, editSql, generateSql, titleSql } from './sql'
-import { chatRlsPolicy } from './sql.edge'
+import { chatRlsPolicy, generateV2 } from './sql.edge'
 
-const openAiKey = process.env.OPENAI_KEY
-const openai = new OpenAI({ apiKey: openAiKey })
+const { openai, model, error } = createOpenAiClient()
+
+if (error) {
+  throw error
+}
+
+console.info(`Running AI tests against model '${model}'`)
 
 describe('generate', () => {
   test('single table with specified columns', async () => {
     const { sql, title } = await generateSql(
       openai,
+      model,
       'create a table to track employees with name, email, and position'
     )
 
@@ -24,6 +30,7 @@ describe('edit', () => {
   test('add length constraint', async () => {
     const { sql } = await editSql(
       openai,
+      model,
       'force name to be at least 4 characters',
       codeBlock`
         create table employees (
@@ -42,6 +49,7 @@ describe('debug', () => {
   test('fix order of operations', async () => {
     const { sql } = await debugSql(
       openai,
+      model,
       'relation "departments" does not exist',
       codeBlock`
         create table employees (
@@ -64,6 +72,7 @@ describe('debug', () => {
   test('fix typos', async () => {
     const { sql, solution } = await debugSql(
       openai,
+      model,
       'syntax error at or near "fromm"',
       codeBlock`
         select * fromm employees;
@@ -79,6 +88,7 @@ describe('title', () => {
   test('title matches content', async () => {
     const { title, description } = await titleSql(
       openai,
+      model,
       codeBlock`
         create table employees (
           id bigint primary key generated always as identity,
@@ -103,6 +113,7 @@ describe('rls chat', () => {
   test('select policy using table definition', async () => {
     const responseStream = await chatRlsPolicy(
       openai,
+      model,
       [
         {
           role: 'user',
@@ -131,3 +142,32 @@ describe('rls chat', () => {
     )
   })
 })
+
+describe('sql chat', () => {
+  test('single table with specified columns', async () => {
+    const responseStream = await generateV2(openai, model, [
+      {
+        role: 'user',
+        content: 'create a table to track employees with name, email, and position',
+      },
+    ])
+    const responseText = await collectStream(responseStream)
+    const [sql] = extractMarkdownSql(responseText)
+
+    expect(sql).toMatchSnapshot()
+  })
+
+  test('table referencing users creates fk to auth.users', async () => {
+    const responseStream = await generateV2(openai, model, [
+      {
+        role: 'user',
+        content: 'create a todos table and track author',
+      },
+    ])
+
+    const responseText = await collectStream(responseStream)
+    const [sql] = extractMarkdownSql(responseText)
+
+    expect(sql).toMatchSnapshot()
+  })
+})
diff --git a/packages/ai-commands/src/sql.ts b/packages/ai-commands/src/sql.ts
index bd39f633c9a..2043c00ec6d 100644
--- a/packages/ai-commands/src/sql.ts
+++ b/packages/ai-commands/src/sql.ts
@@ -9,13 +9,6 @@ const generateSqlSchema = SchemaBuilder.emptySchema()
   .addString('sql', {
     description: stripIndent`
       The generated SQL (must be valid SQL).
-      - For primary keys, always use "id bigint primary key generated always as identity" (not serial)
-      - Prefer creating foreign key references in the create statement
-      - Prefer 'text' over 'varchar'
-      - Prefer 'timestamp with time zone' over 'date'
-      - Use vector(384) data type for any embedding/vector related query
-      - Always use double apostrophe in SQL strings (eg. 'Night''s watch')
-      - Always use semicolons
     `,
   })
   .addString('title', {
@@ -28,16 +21,6 @@ const generateSqlSchema = SchemaBuilder.emptySchema()
 const editSqlSchema = SchemaBuilder.emptySchema().addString('sql', {
   description: stripIndent`
       The modified SQL (must be valid SQL).
-      - Assume the query hasn't been executed yet
-      - For primary keys, always use "id bigint primary key generated always as identity" (not serial)
-      - When creating tables, always add foreign key references inline
-      - Prefer 'text' over 'varchar'
-      - Prefer 'timestamp with time zone' over 'date'
-      - Use vector(384) data type for any embedding/vector related query
-      - Always use double apostrophe in SQL strings (eg. 'Night''s watch')
-      - Use real examples when possible
-      - Add constraints if requested
-      - Always use semicolons
     `,
 })
 
@@ -83,28 +66,12 @@ const completionFunctions = {
   },
   debugSql: {
     name: 'debugSql',
-    description: stripIndent`
-      Debugs a Postgres SQL error. Returns the fixed SQL and a solution explaining it.
-      - Create extensions if they are missing (only for valid extensions)
-      - Suggest creating tables if they are missing
-      - Include all of the original SQL
-      - For primary keys, always use "id bigint primary key generated always as identity" (not serial)
-      - When creating tables, always add foreign key references inline
-      - Prefer 'text' over 'varchar'
-      - Prefer 'timestamp with time zone' over 'date'
-      - Use vector(384) data type for any embedding/vector related query
-      - Always use double apostrophe in SQL strings (eg. 'Night''s watch')
-      - Always use semicolons
-    `,
+    description: 'Debugs a Postgres SQL error. Returns the fixed SQL and a solution explaining it.',
     parameters: debugSqlSchema.schema as Record<string, unknown>,
   },
   generateTitle: {
     name: 'generateTitle',
-    description: stripIndent`
-      Generates a short title and summarized description for a Postgres SQL snippet.
-      
-      The description should describe why this table was created (eg. "Table to track todos")
-    `,
+    description: 'Generates a short title and summarized description for a Postgres SQL snippet.',
     parameters: generateTitleSchema.schema as Record<string, unknown>,
   },
 } satisfies Record<string, OpenAI.Chat.Completions.ChatCompletionCreateParams.Function>
@@ -114,11 +81,36 @@ const completionFunctions = {
  *
  * @returns The generated SQL along with a title for it.
  */
-export async function generateSql(openai: OpenAI, prompt: string, entityDefinitions?: string[]) {
+export async function generateSql(
+  openai: OpenAI,
+  model: string,
+  prompt: string,
+  entityDefinitions?: string[]
+) {
   const hasEntityDefinitions = entityDefinitions !== undefined && entityDefinitions.length > 0
 
   const completionMessages: OpenAI.Chat.Completions.ChatCompletionMessageParam[] = []
 
+  completionMessages.push({
+    role: 'system',
+    content: stripIndent`
+      You are a Postgres SQL generator. You must generate Postgres SQL based on a natural language prompt
+
+      The generated SQL (must be valid SQL).
+        - For primary keys, always use "id bigint primary key generated always as identity" (not serial)
+        - Prefer creating foreign key references in the create statement
+        - When creating new tables, always add foreign key references inline (ie. "my_column my_type references ..." on a single line)
+        - Include all necessary tables and create them in the correct order (eg. book references author, so create author before book)
+        - Prefer 'text' over 'varchar'
+        - Prefer 'timestamp with time zone' over 'date'
+        - Use vector(384) data type for any embedding/vector related query
+        - Always use double apostrophe in SQL strings (eg. 'Night''s watch')
+        - Always use semicolons
+
+        Response must contain only JSON satisfying the 'tool_calls'.
+    `,
+  })
+
   if (hasEntityDefinitions) {
     completionMessages.push({
       role: 'user',
@@ -136,7 +128,7 @@ export async function generateSql(openai: OpenAI, prompt: string, entityDefiniti
 
   try {
     const completionResponse = await openai.chat.completions.create({
-      model: 'gpt-3.5-turbo-0125',
+      model,
       messages: completionMessages,
       max_tokens: 1024,
       temperature: 0,
@@ -188,6 +180,7 @@ export async function generateSql(openai: OpenAI, prompt: string, entityDefiniti
  */
 export async function editSql(
   openai: OpenAI,
+  model: string,
   prompt: string,
   sql: string,
   entityDefinitions?: string[]
@@ -196,6 +189,27 @@ export async function editSql(
 
   const completionMessages: OpenAI.Chat.Completions.ChatCompletionMessageParam[] = []
 
+  completionMessages.push({
+    role: 'system',
+    content: stripIndent`
+      You are a Postgres SQL editor. You must edit Postgres SQL based on the user's instructions.
+
+      - Assume the query hasn't been executed yet
+      - For primary keys, always use "id bigint primary key generated always as identity" (not serial)
+      - When creating new tables, always add foreign key references inline (ie. "my_column my_type references ..." on a single line)
+      - Include all necessary tables and create them in the correct order (eg. book references author, so create author before book)
+      - Prefer 'text' over 'varchar'
+      - Prefer 'timestamp with time zone' over 'date'
+      - Use vector(384) data type for any embedding/vector related query
+      - Always use double apostrophe in SQL strings (eg. 'Night''s watch')
+      - Use real examples when possible
+      - Add constraints if requested
+      - Always use semicolons
+
+      Response must contain only JSON satisfying the 'tool_calls'.
+    `,
+  })
+
   if (hasEntityDefinitions) {
     completionMessages.push({
       role: 'user',
@@ -222,7 +236,7 @@ export async function editSql(
 
   try {
     const completionResponse = await openai.chat.completions.create({
-      model: 'gpt-3.5-turbo-0125',
+      model,
       messages: completionMessages,
       max_tokens: 2048,
       temperature: 0,
@@ -274,6 +288,7 @@ export async function editSql(
  */
 export async function debugSql(
   openai: OpenAI,
+  model: string,
   errorMessage: string,
   sql: string,
   entityDefinitions?: string[]
@@ -282,6 +297,25 @@ export async function debugSql(
 
   const completionMessages: OpenAI.Chat.Completions.ChatCompletionMessageParam[] = []
 
+  completionMessages.push({
+    role: 'system',
+    content: stripIndent`
+      You are a Postgres SQL debugger. You must fix the user's SQL and a provide solution explaining it.
+
+      - Create extensions if they are missing (only for valid extensions)
+      - Suggest creating tables if they are missing
+      - Include all of the original SQL
+      - For primary keys, always use "id bigint primary key generated always as identity" (not serial)
+      - When creating tables, always add foreign key references inline
+      - Prefer 'text' over 'varchar'
+      - Prefer 'timestamp with time zone' over 'date'
+      - Use vector(384) data type for any embedding/vector related query
+      - Always use double apostrophe in SQL strings (eg. 'Night''s watch')
+      - Always use semicolons
+      - Do not mix up the user's error with "tool use failed"
+    `,
+  })
+
   if (hasEntityDefinitions) {
     completionMessages.push({
       role: 'user',
@@ -311,7 +345,7 @@ export async function debugSql(
 
   try {
     const completionResponse = await openai.chat.completions.create({
-      model: 'gpt-3.5-turbo-0125',
+      model,
       messages: completionMessages,
       max_tokens: 2048,
       temperature: 0,
@@ -361,17 +395,29 @@ export async function debugSql(
  *
  * @returns A title and description for the SQL snippet.
  */
-export async function titleSql(openai: OpenAI, sql: string) {
-  const completionMessages: OpenAI.Chat.Completions.ChatCompletionMessageParam[] = [
-    {
-      role: 'user',
-      content: sql,
-    },
-  ]
+export async function titleSql(openai: OpenAI, model: string, sql: string) {
+  const completionMessages: OpenAI.Chat.Completions.ChatCompletionMessageParam[] = []
+
+  completionMessages.push({
+    role: 'system',
+    content: stripIndent`
+      You are a Postgres SQL title/description generator.
+      You must generates a short title and summarized description for a Postgres SQL snippet.
+
+      - The title should be short but still address all parts of the code.
+      - The description should describe why this table was created (eg. "Table to track todos")
+      while addressing all parts of the code.
+    `,
+  })
+
+  completionMessages.push({
+    role: 'user',
+    content: sql,
+  })
 
   try {
     const completionResponse = await openai.chat.completions.create({
-      model: 'gpt-3.5-turbo-0125',
+      model,
       messages: completionMessages,
       max_tokens: 1024,
       temperature: 0,
diff --git a/packages/ai-commands/src/tokenizer.test.ts b/packages/ai-commands/src/tokenizer.test.ts
new file mode 100644
index 00000000000..6ae5fffa775
--- /dev/null
+++ b/packages/ai-commands/src/tokenizer.test.ts
@@ -0,0 +1,24 @@
+import { describe, expect, test } from '@jest/globals'
+import { Message, tokenizeChat } from './tokenizer'
+
+const sampleMessages: Message[] = [
+  { role: 'system', content: 'You are a helpful Postgres SQL assistant' },
+  { role: 'user', content: 'Help me write SQL to get all movies that start with T' },
+  {
+    role: 'assistant',
+    content: "Sure!\n```sql\nselect * from movies where title like 'T%';\n```",
+  },
+  { role: 'user', content: 'Thanks! What about S?' },
+]
+
+describe('tokenizer', () => {
+  test('counts llama3 tokens correctly', async () => {
+    const tokens = await tokenizeChat(sampleMessages, 'llama3')
+    expect(tokens.length).toBe(68)
+  })
+
+  test('counts gpt-3.5-turbo tokens correctly', async () => {
+    const tokens = await tokenizeChat(sampleMessages, 'gpt-3.5-turbo')
+    expect(tokens.length).toBe(66)
+  })
+})
diff --git a/packages/ai-commands/src/tokenizer.ts b/packages/ai-commands/src/tokenizer.ts
index b00be5ab75f..6832b6a7f01 100644
--- a/packages/ai-commands/src/tokenizer.ts
+++ b/packages/ai-commands/src/tokenizer.ts
@@ -1,88 +1,191 @@
-import { getEncoding } from 'js-tiktoken'
-import OpenAI from 'openai'
+import { AutoTokenizer, PreTrainedTokenizer } from '@xenova/transformers'
 
-export const tokenizer = getEncoding('cl100k_base')
+export interface Message {
+  role: 'system' | 'user' | 'assistant'
+  content: string
+}
 
-/**
- * Count the tokens for multi-message chat completion requests
- */
-export function getChatRequestTokenCount(
-  messages: OpenAI.Chat.Completions.ChatCompletionMessageParam[],
-  model = 'gpt-3.5-turbo-0301'
-): number {
-  const tokensPerRequest = 3 // every reply is primed with <|im_start|>assistant<|im_sep|>
-  const numTokens = messages.reduce((acc, message) => acc + getMessageTokenCount(message, model), 0)
-
-  return numTokens + tokensPerRequest
+function isNumberArray(value: any): value is number[] {
+  return Array.isArray(value) && typeof value[0] === 'number'
 }
 
 /**
- * Count the tokens for a single message within a chat completion request
- *
- * See "Counting tokens for chat API calls"
- * from https://github.com/openai/openai-cookbook/blob/834181d5739740eb8380096dac7056c925578d9a/examples/How_to_count_tokens_with_tiktoken.ipynb
+ * Gets the Hugging Face model path for the given model.
  */
-export function getMessageTokenCount(
-  message: OpenAI.Chat.Completions.ChatCompletionMessageParam,
-  model = 'gpt-3.5-turbo-0301'
-): number {
-  let tokensPerMessage: number
-  let tokensPerName: number
-
+function getModelPath(model: string) {
   switch (model) {
-    case 'gpt-3.5-turbo':
-      console.warn(
-        'Warning: gpt-3.5-turbo may change over time. Returning num tokens assuming gpt-3.5-turbo-0301.'
-      )
-      return getMessageTokenCount(message, 'gpt-3.5-turbo-0301')
+    case 'llama3':
+    case 'llama3-70b-8192':
+    case 'llama3-8b-8192':
+      return 'Xenova/llama-3-tokenizer'
+    case 'gpt-4-turbo':
+    case 'gpt-4-turbo-preview':
+    case 'gpt-4-turbo-2024-04-09':
+    case 'gpt-4-0125-preview':
+    case 'gpt-4-1106-preview':
+    case 'gpt-4-vision-preview':
+    case 'gpt-4-1106-vision-preview':
+    case 'gpt-4-32k':
+    case 'gpt-4-32k-0613':
     case 'gpt-4':
-      console.warn('Warning: gpt-4 may change over time. Returning num tokens assuming gpt-4-0314.')
-      return getMessageTokenCount(message, 'gpt-4-0314')
-    case 'gpt-3.5-turbo-0301':
-      tokensPerMessage = 4 // every message follows <|start|>{role/name}\n{content}<|end|>\n
-      tokensPerName = -1 // if there's a name, the role is omitted
-      break
+    case 'gpt-4-0613':
     case 'gpt-4-0314':
-      tokensPerMessage = 3
-      tokensPerName = 1
-      break
-    default:
-      throw new Error(
-        `Unknown model '${model}'. See https://github.com/openai/openai-python/blob/main/chatml.md for information on how messages are converted to tokens.`
-      )
-  }
-
-  return Object.entries(message).reduce((acc, [key, value]) => {
-    acc += tokenizer.encode(value).length
-    if (key === 'name') {
-      acc += tokensPerName
-    }
-    return acc
-  }, tokensPerMessage)
-}
-
-/**
- * Get the maximum number of tokens for a model's context.
- *
- * Includes tokens in both message and completion.
- */
-export function getMaxTokenCount(model: string): number {
-  switch (model) {
     case 'gpt-3.5-turbo':
-      console.warn(
-        'Warning: gpt-3.5-turbo may change over time. Returning max num tokens assuming gpt-3.5-turbo-0301.'
-      )
-      return getMaxTokenCount('gpt-3.5-turbo-0301')
-    case 'gpt-4':
-      console.warn(
-        'Warning: gpt-4 may change over time. Returning max num tokens assuming gpt-4-0314.'
-      )
-      return getMaxTokenCount('gpt-4-0314')
+    case 'gpt-3.5-turbo-0125':
+    case 'gpt-3.5-turbo-1106':
+    case 'gpt-3.5-turbo-16k':
+    case 'gpt-3.5-turbo-16k-0613':
+    case 'gpt-3.5-turbo-instruct':
+    case 'gpt-3.5-turbo-0613':
     case 'gpt-3.5-turbo-0301':
-      return 4097
-    case 'gpt-4-0314':
-      return 4097
+      return 'Xenova/gpt-4'
     default:
       throw new Error(`Unknown model '${model}'`)
   }
 }
+
+const tokenizers = new Map<string, PreTrainedTokenizer>()
+
+/**
+ * Creates a tokenizer for the specified model using
+ * Transformers.js and config downloaded from Hugging Face.
+ *
+ * Caches the tokenizer in memory for subsequent uses.
+ */
+async function getTokenizer(model: string) {
+  const cachedTokenizer = tokenizers.get(model)
+
+  if (cachedTokenizer) {
+    return cachedTokenizer
+  }
+
+  const modelPath = getModelPath(model)
+  const tokenizer = await AutoTokenizer.from_pretrained(modelPath)
+
+  tokenizers.set(model, tokenizer)
+
+  return tokenizer
+}
+
+/**
+ * Tokenizes plain text for the specified model.
+ */
+export async function tokenize(content: string, model: string) {
+  const tokenizer = await getTokenizer(model)
+
+  const tokens = tokenizer.encode(content)
+
+  return tokens
+}
+
+/**
+ * Tokenizes chat messages for the specified model.
+ *
+ * Includes delimiter tokens like <|im_start|>, <|im_end|>, etc
+ */
+export async function tokenizeChat(messages: Message[], model: string) {
+  const tokenizer = await getTokenizer(model)
+
+  const tokens = tokenizer.apply_chat_template(messages, {
+    tokenize: true,
+    return_tensor: false,
+    add_generation_prompt: true,
+  })
+
+  if (!isNumberArray(tokens)) {
+    throw new Error(`Unexpected tokenizer output. Expected number array.`)
+  }
+
+  return tokens
+}
+
+/**
+ * Counts the number of tokens in plain text for the specified model.
+ */
+export async function countTokens(content: string, model: string) {
+  const tokens = await tokenize(content, model)
+
+  return tokens.length
+}
+
+/**
+ * Counts the number of tokens in the chat messages for the specified model.
+ *
+ * Accounts for delimiter tokens like <|im_start|>, <|im_end|>, etc
+ */
+export async function countChatTokens(messages: Message[], model: string) {
+  const tokens = await tokenizeChat(messages, model)
+
+  return tokens.length
+}
+
+/**
+ * Get the maximum number of tokens in a model's context window.
+ *
+ * Defaults to 4096 for unknown models.
+ */
+export function getContextWindow(model: string): number {
+  switch (model) {
+    case 'gpt-4-turbo':
+    case 'gpt-4-turbo-preview':
+    case 'gpt-4-turbo-2024-04-09':
+    case 'gpt-4-0125-preview':
+    case 'gpt-4-1106-preview':
+    case 'gpt-4-vision-preview':
+    case 'gpt-4-1106-vision-preview':
+      return 128000
+    case 'gpt-4-32k':
+    case 'gpt-4-32k-0613':
+      return 32768
+    case 'gpt-3.5-turbo':
+    case 'gpt-3.5-turbo-0125':
+    case 'gpt-3.5-turbo-1106':
+    case 'gpt-3.5-turbo-16k':
+    case 'gpt-3.5-turbo-16k-0613':
+      return 16385
+    case 'gpt-4':
+    case 'gpt-4-0613':
+    case 'llama3':
+    case 'llama3-70b-8192':
+    case 'llama3-8b-8192':
+      return 8192
+    case 'gpt-4-0314':
+    case 'gpt-3.5-turbo-instruct':
+    case 'gpt-3.5-turbo-0613':
+    case 'gpt-3.5-turbo-0301':
+    default:
+      return 4096
+  }
+}
+
+/**
+ * Remove conversation messages as needed until the entire
+ * request fits the context window of the specified model.
+ *
+ * @param systemMessages The initial system messages that must always exist
+ * @param conversationMessages The conversation messages that can be trimmed if necessary
+ * @param model The model that will be responding to these messages
+ * @param maxCompletionTokens The number of response tokens to leave room for
+ *
+ * @returns The final set of messages that will fit in the context window of the model.
+ */
+export async function capMessages(
+  systemMessages: Message[],
+  conversationMessages: Message[],
+  model: string,
+  maxCompletionTokens = 0
+) {
+  const contextWindow = getContextWindow(model)
+  const trimmedConversationMessages = [...conversationMessages]
+
+  let tokenCount = await countChatTokens([...systemMessages, ...conversationMessages], model)
+
+  // Remove earlier conversation messages until we fit
+  while (tokenCount + maxCompletionTokens >= contextWindow) {
+    trimmedConversationMessages.shift()
+
+    tokenCount = await countChatTokens([...systemMessages, ...conversationMessages], model)
+  }
+
+  return [...systemMessages, ...trimmedConversationMessages]
+}
diff --git a/packages/ai-commands/test/extensions.ts b/packages/ai-commands/test/extensions.ts
index bcce558937f..939a3f7474a 100644
--- a/packages/ai-commands/test/extensions.ts
+++ b/packages/ai-commands/test/extensions.ts
@@ -1,14 +1,15 @@
 import { expect } from '@jest/globals'
 import { codeBlock } from 'common-tags'
-import OpenAI from 'openai'
+import { createOpenAiClient } from '../src/openai'
 
-const openAiKey = process.env.OPENAI_KEY
-const openai = new OpenAI({ apiKey: openAiKey })
+const { openai, model, error } = createOpenAiClient()
+
+if (error) {
+  throw error
+}
 
 expect.extend({
   async toMatchCriteria(received: string, criteria: string) {
-    const model = 'gpt-4-1106-preview'
-
     const completionResponse = await openai.chat.completions.create({
       model,
       messages: [
@@ -21,7 +22,9 @@ expect.extend({
             - \`{ "pass": true, "reason": "<reason>" }\` if 'Received' adheres to the test 'Criteria'
             - \`{ "pass": false, "reason": "<reason>" }\` if 'Received' does not adhere to the test 'Criteria'
 
+            Be critical and make sure every part of 'Criteria' is addressed in 'Received'.
             The "reason" must explain exactly which part of 'Received' did or did not pass the test 'Criteria'.
+            Be detailed as to why you failed the test.
           `,
         },
         {