Files
supabase/apps/studio/pages/api/ai/code/complete.ts
T
Matt Rossman adf8b0c67c feat(assistant): per-endpoint reasoningEffort + model config cleanup (#43981)
We're exploring support for newer models like
[gpt-5.4-nano](https://openai.com/index/introducing-gpt-5-4-mini-and-nano/)
in Assistant. This model doesn't support the `'minimal'` reasoning
effort level we use for gpt-5-mini which leads to vague errors.

<img width="595" height="263" alt="CleanShot 2026-03-18 at 17 13 05@2x"
src="https://github.com/user-attachments/assets/cf7c2370-322d-4a8a-be55-23e680db0aa0"
/>


Also, we've [previously
discussed](https://supabase.slack.com/archives/C0161K73J1J/p1771544464850199?thread_ts=1771493920.775699&cid=C0161K73J1J)
that reasoning adds unnecessary latency to otherwise simple AI
completion endpoints like `title-v2`. We want more control of reasoning
level independent of model/endpoint.

This PR aims to solve both problems by:
- making reasoning effort configurable on a per-request basis
- adding compile-time guardrails to prevent selecting an incompatible
reasoning level for models
- adding a `DEFAULT_COMPLETION_MODEL` with minimal reasoning that we can
update with newer models that support disabling reasoning (independent
of Assistant chat model reasoning)

Other improvements to our model config logic:
- Fixes bug in `onboarding/design.ts` and `assistant.eval.ts` where
`providerOptions` was being dropped
- `getModel()` now returns a bundled `modelParams` object (spread into
AI SDK calls) so `providerOptions` can't be accidentally omitted (this
[has happened
before](https://supabase.slack.com/archives/C0161K73J1J/p1771518443534309?thread_ts=1771493920.775699&cid=C0161K73J1J))
- Introduces an `ASSISTANT_MODELS` registry as a single source of truth
for assistant model config, eliminating hardcoded model IDs across the
codebase
- Aligns free/pro model conditional logic with `assistant.advance_model`
entitlement naming conventions instead of the `isLimited` pattern
- Adds `console.error` logging of Assistant stream errors so we can
interpret reasoning effort compatibility errors in the future (instead
of just opaque "Sorry, I'm having trouble responding right now" card)
- Removes unnecessary type casts and generally making the model config
logic stricter
- Removes pre-existing dead code: `anthropic` provider variant in
`GetModelParams` / `PROVIDERS` registry that was never implemented in
`getModel()`

Now if you try to select an unsupported reasoning level you get a type
error:

<img width="1306" height="320" alt="CleanShot 2026-03-20 at 14 37 24@2x"
src="https://github.com/user-attachments/assets/a6ac234b-5ea5-4d81-8e01-ac4be34a0800"
/>

And if for some reason an invalid reasoning level slips through, you now
get a server-side error surfacing the issue:

<img width="1268" height="204" alt="CleanShot 2026-03-20 at 14 58 14@2x"
src="https://github.com/user-attachments/assets/aadc1b7a-9495-475f-9741-39979bd27cd7"
/>

I've tested gpt-5 and gpt-5-mini are still working on the staging
preview and verified the models were selected properly in Braintrust
logs. Both models are available on my Pro test account, and my Free test
account shows the Pro upgrade CTA.


Closes AI-446
Closes AI-551
2026-03-25 11:29:23 -04:00

177 lines
5.6 KiB
TypeScript

import pgMeta from '@supabase/pg-meta'
import { generateText, ModelMessage, stepCountIs } from 'ai'
import { IS_PLATFORM } from 'common'
import { source } from 'common-tags'
import { executeSql } from 'data/sql/execute-sql-query'
import { AiOptInLevel } from 'hooks/misc/useOrgOptedIntoAi'
import { getModel } from 'lib/ai/model'
import { DEFAULT_COMPLETION_MODEL } from 'lib/ai/model.utils'
import { getOrgAIDetails } from 'lib/ai/org-ai-details'
import {
EDGE_FUNCTION_PROMPT,
GENERAL_PROMPT,
OUTPUT_ONLY_PROMPT,
PG_BEST_PRACTICES,
RLS_PROMPT,
SECURITY_PROMPT,
} from 'lib/ai/prompts'
import { getTools } from 'lib/ai/tools'
import apiWrapper from 'lib/api/apiWrapper'
import { executeQuery } from 'lib/api/self-hosted/query'
import { NextApiRequest, NextApiResponse } from 'next'
import z from 'zod'
export const maxDuration = 60
async function handler(req: NextApiRequest, res: NextApiResponse) {
if (req.method !== 'POST') {
return new Response(
JSON.stringify({ data: null, error: { message: `Method ${req.method} Not Allowed` } }),
{
status: 405,
headers: { 'Content-Type': 'application/json', Allow: 'POST' },
}
)
}
try {
const { completionMetadata, projectRef, connectionString, orgSlug, language } = req.body
const { textBeforeCursor, textAfterCursor, prompt, selection } = completionMetadata
if (!projectRef) {
return res.status(400).json({
error: 'Missing project_ref in request body',
})
}
const authorization = req.headers.authorization
const accessToken = authorization?.replace('Bearer ', '')
let aiOptInLevel: AiOptInLevel = 'disabled'
if (!IS_PLATFORM) {
aiOptInLevel = 'schema'
}
if (IS_PLATFORM && orgSlug && authorization && projectRef) {
// Get organizations and compute opt in level server-side
const { aiOptInLevel: orgAIOptInLevel } = await getOrgAIDetails({
orgSlug,
authorization,
projectRef,
})
aiOptInLevel = orgAIOptInLevel
}
// For code completion, we always use the limited model
const {
modelParams,
error: modelError,
promptProviderOptions,
} = await getModel({
provider: 'openai',
modelEntry: DEFAULT_COMPLETION_MODEL,
})
if (modelError) {
return res.status(500).json({ error: modelError.message })
}
// Get a list of all schemas to add to context
const pgMetaSchemasList = pgMeta.schemas.list()
type Schemas = z.infer<(typeof pgMetaSchemasList)['zod']>
const { result: schemas } =
aiOptInLevel !== 'disabled'
? await executeSql<Schemas>(
{
projectRef,
connectionString,
sql: pgMetaSchemasList.sql,
},
undefined,
{
'Content-Type': 'application/json',
...(authorization && { Authorization: authorization }),
},
IS_PLATFORM ? undefined : executeQuery
)
: { result: [] }
const schemasString =
schemas?.length > 0
? `The available database schema names are: ${JSON.stringify(schemas)}`
: "You don't have access to any schemas."
// Important: do not use dynamic content in the system prompt or Bedrock will not cache it
const system = source`
${GENERAL_PROMPT}
${OUTPUT_ONLY_PROMPT}
${language === 'sql' ? PG_BEST_PRACTICES : EDGE_FUNCTION_PROMPT}
${language === 'sql' && RLS_PROMPT}
${SECURITY_PROMPT}
`
// Note: these must be of type `CoreMessage` to prevent AI SDK from stripping `providerOptions`
// https://github.com/vercel/ai/blob/81ef2511311e8af34d75e37fc8204a82e775e8c3/packages/ai/core/prompt/standardize-prompt.ts#L83-L88
const coreMessages: ModelMessage[] = [
{
role: 'system',
content: system,
...(promptProviderOptions && { providerOptions: promptProviderOptions }),
},
{
role: 'assistant',
// Add any dynamic context here
content: `
You are helping me edit some code.
Here is the context:
${textBeforeCursor}<selection>${selection}</selection>${textAfterCursor}
The available database schema names are: ${schemasString}
Instructions:
1. Only modify the selected text based on this prompt: ${prompt}
2. Your response should be ONLY the modified selection text, nothing else. Remove selected text if needed.
3. Do not wrap in code blocks or markdown
4. You can respond with one word or multiple words
5. Ensure the modified text flows naturally within the current line
6. Avoid duplicating code when considering the full statement
7. If there is no surrounding context (before or after), make sure your response is a complete valid SQL statement that can be run and resolves the prompt.
Modify the selected text now:
`,
},
]
// Get tools
const tools = await getTools({
projectRef,
connectionString,
authorization,
aiOptInLevel,
accessToken,
})
const { text } = await generateText({
...modelParams,
stopWhen: stepCountIs(5),
messages: coreMessages,
tools,
})
return res.status(200).json(text)
} catch (error) {
console.error('Completion error:', error)
return res.status(500).json({
error: 'Failed to generate completion',
})
}
}
const wrapper = (req: NextApiRequest, res: NextApiResponse) =>
apiWrapper(req, res, handler, { withAuth: true })
export default wrapper