Files
supabase/apps/studio/pages/api/ai/sql/generate-v4.ts
Matt Rossman d143571586 feat(assistant): trace-level scorers + server-side tool execution with needsApproval (#45654)
## Motivation

When Assistant runs a potentially destructive tool like `execute_sql`,
it stops the LLM request and prompts for client-side approval and
execution of the tool. After approval, a second request kicks off under
a separate trace. This has made scoring and
[Topics](https://www.braintrust.dev/blog/topics) classification
challenging, as the generated `output` is split across stateless
requests. The [span-level
scoring](https://www.braintrust.dev/docs/evaluate/custom-code#score-spans)
approach we've used thusfar (after the LLM call, we massage the result
into an `output` payload that's stuck onto the root span) has been
cumbersome and led to invalid scores / topics where only part of the
assistant response is considered. It's also inefficient, as we're
duplicating potentially large info (like the `search_docs` output) that
already exists within the trace.

An alternative to scoring spans is to [score
traces](https://www.braintrust.dev/docs/evaluate/custom-code#score-traces).
Braintrust [best
practices](https://www.braintrust.dev/docs/evaluate/score-online#best-practices)
advise:

> Use span scope for evaluating individual operations or outputs. Use
trace scope for evaluating multi-turn conversations, overall workflow
completion, or when your scorer needs access to the full execution
context.

We've also received [direct
guidance](https://supabase.slack.com/archives/C05QYJBLX89/p1777925770927149?thread_ts=1777905716.911979&cid=C05QYJBLX89)
from their team to use this approach.

## Changes

Migrates eval scorers from custom `AssistantEvalOutput` shape to
trace-level scoring via `trace.getThread()` / `trace.getSpans()`, with
thread parsing that scores the full latest Assistant turn and passes
prior conversation separately where relevant.

Moves `execute_sql` and `deploy_edge_function` from client-side
execution after approval to AI SDK `needsApproval` + server-side
`execute()`. SQL results returned to the model are gated by AI opt-in
level, so row data is only included with `schema_and_log_and_data`;
otherwise the tool returns the no-data-permissions sentinel.

Adds `metadata.isFinalStep` to disambiguate multiple LLM requests within
an "assistant" turn due to tool call requests/responses. For online
evals, this means we should configure automations to only score traces
with `metadata.isFinalStep = true` to ensure we're judging the complete
generated response.

Other minor kaizen changes:
- Renamed `promptProviderOptions` to `systemProviderOptions` to clarify
that this is associated with the "system" message and disambiguate from
the root `providerOptions`
- Adds `evals/trace-utils.ts` to handle Zod validation of the `unknown`
span shapes from Braintrust, to more easily access typed inputs/output
on tool spans.
- Bumps AI SDK floor version `^6.0.116` → `^6.0.174`
- Tweaked the "Conciseness" scorer to not unfairly dock points for the
new `[called tool_name]` labels in serialized assistant response

## Verification

In the studio staging build, I asked Assistant to create a todos table
with 3 sample todos. I manually approved the `execute_sql` call and saw
Assistant generate text before & after the call.

In Braintrust I verified two traces were produced (see [filtered
logs](https://www.braintrust.dev/app/supabase.io/p/Assistant/logs?v=Staging&tvt=trace&search={%22filter%22:[{%22text%22:%22metadata.environment%2520%253D%2520%27staging%27%22,%22label%22:%22metadata.environment%2520%253D%2520%27staging%27%22,%22originType%22:%22btql%22},{%22text%22:%22%2560Chat%2520ID%2560%2520%253D%2520%25221cb2ac45-e5e7-458c-9da4-3bf6863b8842%2522%22,%22label%22:%22Chat%2520ID%2520equals%25201cb2ac45-e5e7-458c-9da4-3bf6863b8842%22,%22originType%22:%22form%22}]})),
the first with `metadata.isFinalStep = false` and the second with
`metadata.isFinalStep = true`.

In the Braintrust staging scorers, I ran the preview Completeness scorer
on the second trace and verified it sees the complete Assistant response
including markers for tool calls ([link to
trace](https://www.braintrust.dev/app/supabase.io/p/Assistant%20(Staging%20Scorers)/trace?object_type=project_logs&object_id=b5214b62-ad1e-4929-9d5b-40b1daebe948&r=0ed0a4f8-8aff-4a34-bb1d-1df1d88a5070&s=ff9015f8-6bf7-4ab3-83a9-ca4e69e27e82))

<img width="1193" height="960" alt="CleanShot 2026-05-07 at 11 27 10@2x"
src="https://github.com/user-attachments/assets/509d4858-c3a1-4068-986d-3aa4d5617d1a"
/>

I also tested the `deploy_edge_function` workflow and verified it still
prompts for permission and warns on deployment of existing functions.

**References**
- https://www.braintrust.dev/docs/evaluate/custom-code#score-traces
-
https://ai-sdk.dev/docs/ai-sdk-core/tools-and-tool-calling#tool-execution-approval

Supercedes https://github.com/supabase/supabase/pull/45556 and
https://github.com/supabase/supabase/pull/45339

Closes AI-473

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit

* **New Features**
* Tool actions (SQL execution, edge-function deploy) now require
explicit user Approve/Deny before proceeding.

* **Improvements**
* Assistant pauses for approval responses before sending follow-ups,
giving clearer control over risky actions.
  * Deploy/replace flows show confirmation and clearer replace warnings.
* Evaluation/scoring updated to use richer trace data for more accurate
assistant performance signals.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
2026-05-12 15:24:21 -04:00

256 lines
7.2 KiB
TypeScript

import pgMeta from '@supabase/pg-meta'
import type { JwtPayload } from '@supabase/supabase-js'
import { safeValidateUIMessages } from 'ai'
import { IS_PLATFORM } from 'common'
import type { NextApiRequest, NextApiResponse } from 'next'
import z from 'zod'
import { executeSql } from '@/data/sql/execute-sql-query'
import type { AiOptInLevel } from '@/hooks/misc/useOrgOptedIntoAi'
import { getOrgAIDetails, getProjectAIDetails } from '@/lib/ai/ai-details'
import { isTracingAllowed } from '@/lib/ai/braintrust-logger'
import { generateAssistantResponse } from '@/lib/ai/generate-assistant-response'
import { getModel } from '@/lib/ai/model'
import {
DEFAULT_ASSISTANT_ADVANCE_MODEL_ID,
DEFAULT_ASSISTANT_BASE_MODEL_ID,
getAssistantModelEntry,
isAssistantBaseModelId,
isKnownAssistantModelId,
type AssistantModelId,
} from '@/lib/ai/model.utils'
import { getTools } from '@/lib/ai/tools'
import apiWrapper from '@/lib/api/apiWrapper'
import { executeQuery } from '@/lib/api/self-hosted/query'
import { getURL } from '@/lib/helpers'
export const maxDuration = 120
export const config = {
api: {
bodyParser: {
sizeLimit: '5mb',
},
},
}
async function handler(req: NextApiRequest, res: NextApiResponse, claims?: JwtPayload) {
const { method } = req
switch (method) {
case 'POST':
return handlePost(req, res, claims)
default:
res.setHeader('Allow', ['POST'])
res.status(405).json({
data: null,
error: { message: `Method ${method} Not Allowed` },
})
}
}
const wrapper = (req: NextApiRequest, res: NextApiResponse) =>
apiWrapper(req, res, handler, { withAuth: true })
export default wrapper
const requestBodySchema = z.object({
messages: z.array(z.any()),
projectRef: z.string(),
connectionString: z.string(),
schema: z.string().optional(),
table: z.string().optional(),
chatId: z.string().optional(),
chatName: z.string().optional(),
orgSlug: z.string().optional(),
model: z.string().optional(),
})
async function handlePost(req: NextApiRequest, res: NextApiResponse, claims?: JwtPayload) {
const authorization = req.headers.authorization
const accessToken = authorization?.replace('Bearer ', '')
if (IS_PLATFORM && !accessToken) {
return res.status(401).json({ error: 'Authorization token is required' })
}
const userId = claims?.sub
const body = typeof req.body === 'string' ? JSON.parse(req.body) : req.body
const { data, error: parseError } = requestBodySchema.safeParse(body)
if (parseError) {
return res.status(400).json({ error: 'Invalid request body', issues: parseError.issues })
}
const {
messages: rawMessages,
projectRef,
connectionString,
orgSlug,
chatId,
chatName,
model: rawRequestedModel,
} = data
const requestedModel: AssistantModelId | undefined =
rawRequestedModel && isKnownAssistantModelId(rawRequestedModel) ? rawRequestedModel : undefined
const messagesValidation = await safeValidateUIMessages({
messages: rawMessages,
})
if (!messagesValidation.success) {
return res.status(400).json({
error: 'Invalid request body',
message: messagesValidation.error.message,
})
}
const messages = messagesValidation.data
let aiOptInLevel: AiOptInLevel = 'disabled'
let hasAccessToAdvanceModel = false
let orgHasHipaaAddon: boolean | undefined
let projectIsSensitive: boolean | undefined
let projectRegion: string | undefined
let orgId: number | undefined
let planId: string | undefined
if (!IS_PLATFORM) {
aiOptInLevel = 'schema'
hasAccessToAdvanceModel = true
}
if (IS_PLATFORM && orgSlug && authorization && projectRef) {
try {
const [orgDetails, projectDetails] = await Promise.all([
getOrgAIDetails({ orgSlug, authorization }),
getProjectAIDetails({ projectRef, authorization }),
])
aiOptInLevel = orgDetails.aiOptInLevel
hasAccessToAdvanceModel = orgDetails.hasAccessToAdvanceModel
orgHasHipaaAddon = orgDetails.hasHipaaAddon
orgId = orgDetails.orgId
planId = orgDetails.planId
projectIsSensitive = projectDetails.isSensitive
projectRegion = projectDetails.region
} catch (error) {
return res.status(400).json({
error: 'There was an error fetching your organization details',
})
}
}
const envThrottled = process.env.IS_THROTTLED !== 'false'
let effectiveModel: AssistantModelId = requestedModel ?? DEFAULT_ASSISTANT_ADVANCE_MODEL_ID
if (!hasAccessToAdvanceModel || (envThrottled && !isAssistantBaseModelId(effectiveModel))) {
effectiveModel = DEFAULT_ASSISTANT_BASE_MODEL_ID
}
const {
modelParams,
error: modelError,
systemProviderOptions,
} = await getModel({
provider: 'openai',
modelEntry: getAssistantModelEntry(effectiveModel),
})
if (modelError) {
return res.status(500).json({ error: modelError.message })
}
try {
const abortController = new AbortController()
req.on('close', () => abortController.abort())
req.on('aborted', () => abortController.abort())
const tools = await getTools({
projectRef,
connectionString,
authorization,
aiOptInLevel,
accessToken,
baseUrl: getURL(),
})
// Get a list of all schemas to add to context
const getSchemas = async (): Promise<string> => {
const pgMetaSchemasList = pgMeta.schemas.list()
type Schemas = z.infer<(typeof pgMetaSchemasList)['zod']>
const { result: schemas } = await executeSql<Schemas>(
{
projectRef,
connectionString,
sql: pgMetaSchemasList.sql,
},
undefined,
{
'Content-Type': 'application/json',
...(authorization && { Authorization: authorization }),
},
IS_PLATFORM ? undefined : executeQuery
)
return schemas?.length > 0
? `The available database schema names are: ${JSON.stringify(schemas)}`
: "You don't have access to any schemas."
}
const result = await generateAssistantResponse({
messages,
...modelParams,
tools,
aiOptInLevel,
getSchemas: aiOptInLevel !== 'disabled' ? getSchemas : undefined,
projectRef,
chatId,
chatName,
allowTracing: isTracingAllowed({
orgHasHipaaAddon,
projectIsSensitive,
projectRegion,
}),
userId,
orgId,
planId,
requestedModel,
systemProviderOptions,
abortSignal: abortController.signal,
onSpanCreated: (spanId) => {
res.setHeader('x-braintrust-span-id', spanId)
},
})
result.pipeUIMessageStreamToResponse(res, {
sendReasoning: true,
headers: { 'Content-Encoding': 'none' },
onError: (error) => {
console.error('Assistant stream error:', error)
if (error == null) {
return 'unknown error'
}
if (typeof error === 'string') {
return error
}
if (error instanceof Error) {
return error.message
}
return JSON.stringify(error)
},
})
} catch (error) {
console.error('Error in handlePost:', error)
if (error instanceof Error) {
return res.status(500).json({ message: error.message })
}
return res.status(500).json({ message: 'An unexpected error occurred.' })
}
}