## Context
As per PR title - also adjusts the imports for files consuming
`apiWrapper` to remove the default export for `apiWrapper`
Have tested locally by throwing an error in one of the API routes -
verified that the event shows up on Sentry
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **Bug Fixes**
* API errors are now captured in Sentry before returning server error
responses, improving production visibility while keeping endpoint
behavior the same.
* **Tests**
* Added coverage to confirm rejected handler executions are reported to
Sentry and return the expected HTTP 500 JSON payload.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
# Sync AI assistant conversation to Front
## What & why
When a user submits a support ticket, an AI assistant chat opens so they
get help
immediately while waiting for a human agent. This PR mirrors every turn
of that chat into
the Front conversation the support form already created, so the support
team sees the full
context and Front automations (routing, emails, CSAT) can act on it.
Studio holds no Front credentials — it calls the platform endpoints (see
the platform PR)
to do the syncing. The assistant card is gated behind the
`supportAssistantFollowUp`
ConfigCat flag.
## How it works
1. **Submit** — `SupportFormV3` generates a stable `threadRef` (via the
`uuid` package —
`crypto.randomUUID()` is `undefined` in insecure contexts like
non-localhost HTTP and
would throw, silently aborting the submit) and sends it on
`/platform/feedback/send`.
The response returns the Front `conversationId`. Both are stored on
`SubmittedSupportRequest`.
2. **Open chat** — `SupportAssistantSuccessCardContent` opens a chat
seeded with
`supportMetadata` (`threadRef`, `frontConversationId`, subject,
category, severity, …).
The first message is a `<support>…</support>` XML block.
3. **First user message** — the chat is tagged `isSupportChat = true`;
the `onFinish` hook
fires `syncSupportChatToFront`.
4. **Subsequent turns** — each `onFinish` slices the unsynced delta,
strips the XML
metadata block from the seed message, and posts to the platform messages
endpoint.
5. **Escalation / resolve** — the `escalate_to_human` /
`resolve_support_conversation` tools
(and manual **Escalate**/**Resolve** buttons in the assistant input)
flip lifecycle status
via `setSupportLifecycleStatus` → `syncSupportLifecycleToFront`, which
calls the
escalation/resolve endpoints. Front rules act on `ai_support_status`.
The assistant only
resolves after the user explicitly confirms the issue is fixed.
## Key design decisions
- **`threadRef` as the shared key** — one UUID travels as `threadRef` on
submit and as
`chatId` on every sync, so all messages thread into a single Front
conversation.
- **`conversationId` from the form response** — passed to all
sync/lifecycle calls so the
platform skips lazy derivation and PATCHes custom fields directly.
- **Delta-only sync** — `lastSyncedMessageCount` tracks what's been
sent; the boundary is
snapshotted before the async call to avoid skipping messages that arrive
mid-flight.
- **Server-side de-dup** — stable `external_id` (`chatId:msg.id`) means
retries don't
duplicate in Front.
- **Fire-and-forget** — sync failures log to Sentry, never break the
chat; `isSyncing`
resets on rehydration so the next `onFinish` retries the same delta.
Message and lifecycle
syncs use separate guards (`isSyncing` / `isLifecycleSyncing`) so an
in-flight message
sync can't drop an escalate/resolve.
- **Lifecycle queued until the conversation exists** — if a lifecycle
transition is requested
before the initial message sync has returned a `frontConversationId`,
it's stored as
`pendingLifecycleStatus` and flushed once the id is assigned, rather
than dropped.
- **Tools return immediately** — the lifecycle tools return a stub to
the AI SDK; the real
Front call happens in `onFinish`, keeping async I/O out of the tool
execute path.
- **XML seed stripped before sync** — only the user's actual `<message>`
is sent to Front
(or dropped entirely if the form already created the conversation).
## Changes
| Area | File(s) |
| --- | --- |
| Support form state | `SupportForm.state.ts` — `threadRef` /
`frontConversationId` on `SubmittedSupportRequest` |
| Support form submit | `support-ticket-send.ts` — sends `threadRef`,
reads `conversationId` |
| Support form UI | `SupportFormV3.tsx` — generates `threadRef`, stores
`conversationId` |
| AI assistant state | `ai-assistant-state.tsx` — `SupportChatMetadata`,
`setSupportLifecycleStatus`, `onFinish` wiring, tool handling |
| Message sync | `state/ai-chat-front-sync.ts` — delta tracking, message
filtering, initial vs. incremental |
| API data layer | `data/feedback/ai-chat-front-sync.ts` — typed
platform-client wrappers for the three conversation endpoints |
| Support tools | `lib/ai/tools/support-tools.ts` — `escalate_to_human`,
`resolve_support_conversation` |
| Tool integration | `lib/ai/tool-filter.ts`, `tools/index.ts`,
`generate-assistant-response.ts` |
| Success card | `SupportAssistantSuccessCardContent.tsx` — tags chat on
first engagement |
| Assistant panel UI | `AIAssistant.tsx` — Escalate/Resolve buttons,
disabled input on closed chats, support placeholders |
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
## Summary by CodeRabbit
- **New Features**
- Support chats now include “Escalate to human” and “Resolve” actions.
- Support submissions can be associated with a stable Front thread via a
generated `threadRef`, preserving linkage across follow-ups.
- AI assistant responses and input hints adapt when support mode is
active.
- **Bug Fixes**
- Improved support chat state management and lifecycle handling to keep
conversation metadata and message history synchronized more reliably
with Front.
- **Chores**
- Added/updated coverage to reflect the new support-chat state and
syncing behavior.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
---------
Co-authored-by: Alaister Young <10985857+alaister@users.noreply.github.com>
## I have read the
[CONTRIBUTING.md](<https://github.com/supabase/supabase/blob/master/CONTRIBUTING.md>)
file.
YES
## What kind of change does this PR introduce?
Feature / refactor.
## What is the current behavior?
The dashboard assistant runs `@supabase/mcp-server-supabase` in-process
over an in-memory transport (`lib/ai/supabase-mcp.ts`).
## What is the new behavior?
The assistant connects to the **remote MCP server** over HTTP
(`@ai-sdk/mcp`), forwarding the dashboard session token as a bearer. URL
comes from `NEXT_PUBLIC_MCP_URL` with a local-dev fallback;
platform-only, and Nimbus works via the same env var.
* **Tool model unchanged:** UI-controlled `execute_sql` (with
`needsApproval`) and `deploy_edge_function` still come from Studio; the
allowlist (`TOOL_CATEGORY_MAP`) remains the gate keeping the remote's
write tools away from the assistant (`read_only` is defense-in-depth).
* **Attribution:** sends `x-source-name: supabase-studio` (+
`x-source-version`) → logged as `source_name`/`client_name`.
* **Connection lifecycle:** the HTTP client is closed via the request's
`AbortSignal` (tools execute later during streaming); `signal` is
required on `getTools`/`getMcpTools`.
* **Resilience:** a remote-MCP failure degrades to the remaining tools
instead of failing the assistant.
* **Drift protection:** relied-upon tools are typed against `keyof
typeof supabaseMcpToolSchemas`, so a package bump that renames/removes
one fails `pnpm typecheck`; a runtime check also warns if the deployed
server returns fewer tools.
* Adds unit tests for the above.
## Additional context
* Verified end-to-end against a local remote MCP server with a dashboard
token: `initialize` 200, tools listed, a tool executed, client closed
cleanly.
* The remote MCP (mgmt-api) already accepts dashboard session tokens
(GoTrue-JWT auth path) — no backend change needed. `NEXT_PUBLIC_MCP_URL`
must point at each env's `/mcp`.
* `@supabase/mcp-server-supabase` is kept — still used by the
self-hosted `/api/mcp` routes.
Closes
[AI-137](https://linear.app/supabase/issue/AI-137/switch-dashboard-assistant-to-remote-mcp)
## Rollout
* **Rollout:** merges with `USE_REMOTE_MCP` off (in-process); flip it to
`true` per environment (staging → prod → Nimbus) once each one's
prerequisites land.
* **Rollback:** unset `USE_REMOTE_MCP` and redeploy to fall back to the
in-process client — no revert needed.
## Summary by CodeRabbit
* **Bug Fixes**
* Improved AI request handling so tool loading and generation clean up
properly when a request is cancelled or the browser connection closes.
* Added safer fallback behavior when remote tool loading fails, so AI
features can continue with available tools instead of stopping entirely.
* Updated remote tool access to use the current project reference and
preserve the correct access headers.
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **New Features**
* AI tools now connect more reliably to remote services and stop cleanly
when requests end or are canceled.
* Tool loading is more resilient, continuing with available tools if
remote access is unavailable.
* **Bug Fixes**
* Improved cleanup to prevent lingering connections during SQL
generation and policy workflows.
* Added safer handling for remote tool changes and invalid responses.
* **Tests**
* Expanded automated coverage for remote tool setup, cancellation, and
fallback behavior.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
## Context
Just some clean up as I was going through stuff
- `useExecuteSqlQuery` is deprecated and not used at all
- As such `execute-sql-query` is technically irrelevant, the more
relevant file is `execute-sql-mutation`
- Hence opting to consolidate `execute-sql-query` into
`execute-sql-mutation`
- Also removing `ExecuteSqlError` since its just re-exporting the
`ResponseError` type
There's a lot of file changes but its essentially just updating the
importing statements across the files
## 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 -->
Mark provenance of SQL via the branded types SafeSqlFragment and
UntrustedSqlFragment. Only SafeSqlFragment should be executed;
UntrustedSqlFragments require some kind of implicit user approval (show
on screen + user has to click something) before they are promoted to
SafeSqlFragment.
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **New Features**
* Editor and RLS tester show loading states for inferred/generated SQL
and include a dedicated user SQL editor for safer edits.
* **Refactor**
* Platform-wide SQL handling tightened: snippets and AI-generated SQL
are treated as untrusted/display-only until promoted, improving safety
and consistency.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
## Context
Resolves FE-3077
Related discussion: https://github.com/orgs/supabase/discussions/45233
Verifying the correctness of your RLS policies set up has always been a
gap, as highlighted by a number of GitHub discussions like
[here](https://github.com/orgs/supabase/discussions/12269) and
[here](https://github.com/orgs/supabase/discussions/14401). As such,
we're piloting a dedicated UI for RLS testing (using role impersonation
as the base), in which you'll be able to
- Run a SQL query as a user (not logged in / logged in - this is the
role impersonation part)
- See which RLS policies are being evaluated as part of the query
- And hopefully be able to debug which policies are not set up correctly
Changes are currently set as a feature preview - and we'll iterate as we
get feedback from everyone 🙂🙏
<img width="613" height="957" alt="image"
src="https://github.com/user-attachments/assets/83c37f8a-28fc-43b3-b0ff-e28571d8710c"
/>
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **New Features**
* RLS Tester: run queries as anon or authenticated users, view inferred
SQL, per-table policy summaries, and data previews of accessible rows.
* UI preview: new RLS Tester preview card and modal with opt-in toggle;
RLS Tester sheet with role/user selector and query editor.
* SQLEditor: “Explain” tab is always visible.
* **Chores**
* Added supporting API endpoints, background checks for table RLS
status, and a local-storage flag to persist the preview opt-in.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
Removes the temporary killswitch added when Braintrust was onboarded as
a subprocessor, to satisfy the 30-day DPA notice obligation. The window
has elapsed and legal has cleared removal.
Drops the `orgIsDpaSigned` check from `isTracingAllowed`, removes the
extra `/platform/organizations/{slug}/documents/dpa-signed` network hop
from `getOrgAIDetails`, and cleans up all call sites and tests.
Closes AI-596
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **Refactor**
* Simplified AI tracing eligibility logic by removing DPA signing status
checks. Tracing authorization decisions now depend solely on region,
HIPAA addon status, and project sensitivity settings.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
**Changes**
- Extracted tracing conditional to an `isTracingAllowed` helper with
unit tests (the function is simple but sensitive hence the extra testing
precaution)
- Disables Braintrust tracing for projects in EU database regions
(region prefix `eu-`) to address GDPR data residency concerns
- Disables Braintrust tracing for orgs whose owners have signed the
previous DPA, as a stopgap during the 30-day notice period for the
updated DPA that adds Braintrust as a subprocessor
- Refactored `org-ai-details.ts` → `ai-details.ts`, splitting
`getOrgAIDetails` into separate org and project helpers to cleanly scope
the EU-region check at the project level
DPA check uses the newly added `/documents/dpa-signed` endpoint from
https://github.com/supabase/platform/pull/31060. This PR includes
regenerated `api.d.ts` and `platform.d.ts` from running `pnpm codegen`
in `packages/api-types` to get type safety on this new endpoint.
Note tracing is still yet to be activated in production, this is a
preparatory step.
**To verify**
Send a chat message and check for the `x-braintrust-span-id` response
header on `POST /api/ai/sql/generate-v4` — it should be absent for
DPA-signed orgs or EU-region projects, and present otherwise.
<img width="3594" height="1992" alt="CleanShot 2026-04-03 at 14 28
58@2x"
src="https://github.com/user-attachments/assets/4c91d7ad-2604-4531-a78e-dedf41632fa5"
/>
If you have access to the Braintrust dashboard, you can also verify
whether logs are produced or not in the Assistant project there.
Closes AI-570
Closes AI-569
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **New Features**
* Tracks organization DPA signing and detects EU-region projects
* Assistant tracing now follows a combined compliance policy (HIPAA
addon, DPA, project sensitivity, region)
* Added helpers to fetch org and project AI details
* **Documentation**
* Expanded API docs with additional examples and clarified parameter
descriptions
* Added response schemas for subscription preview and document status
* **Tests**
* Added/updated tests covering DPA/region behavior and tracing policy
enforcement
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
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
## I have read the
[CONTRIBUTING.md](https://github.com/supabase/supabase/blob/master/CONTRIBUTING.md)
file.
YES
## What kind of change does this PR introduce?
OpenAI claims to support recursive schemas with $defs/$ref, but in
practice it's unreliable. When Zod's z.lazy() is converted to JSON
Schema, it produces recursive $ref entries that OpenAI's structured
output frequently rejects with errors like "Recursive reference
detected" or "Invalid schema for response_format".
Simplify the AI generation schema since we only support AND and don't
need the recursion because we don't support nesting of groups.
Upgrades `ai` from v5 to v6 and all related packages.
**Package bumps:**
- `ai`: `5.0.52` → `^6.0.116`
- `@ai-sdk/openai`: `2.0.32` → `^3.0.41`
- `@ai-sdk/react`: `2.0.52` → `^3.0.118`
- `@ai-sdk/provider`: `^2.0.0` → `^3.0.8`
- `@ai-sdk/provider-utils`: `^3.0.0` → `^4.0.19`
- `@ai-sdk/amazon-bedrock`: `^3.0.0` → `^4.0.81`
- `@ai-sdk/mcp`: N/A → `^1.0.25`
- `openai`: bumped to `^4.104.0`
- `braintrust`: `3.0.x` → `^3.4.0`
**Breaking change migrations:**
- `generateObject` removed in v6 — migrated 5 API routes to
`generateText` with `Output.object({ schema })`, returning
`result.output`
- `convertToModelMessages` is now async — added `await`
- MCP import path changed: `experimental_createMCPClient` from `ai` →
`createMCPClient` from `@ai-sdk/mcp`
- `openai()` defaults to Responses API — added `store: false` to
provider options for ZDR org compatibility
**Streaming fix:**
Added `Content-Encoding: none` header to `pipeUIMessageStreamToResponse`
calls. Without it, proxy middleware buffers the entire SSE response
before flushing, causing the full reply to appear at once.
**Zero Data Retention fix:**
In recent AI SDK versions, `openai()` default to Responses API instead
of the legacy chat completions API. This produces a 404 from OpenAI with
message `"Items are not persisted for Zero Data Retention organizations.
Remove this item from your input and try again."` The Responses API is
OpenAI's [recommended
endpoint](https://developers.openai.com/api/docs/guides/migrate-to-responses).
This PR adds `store: false` as mentioned in
https://github.com/vercel/ai/issues/10060 to avoid incompatible
persistence attempts.
**References:**
- https://ai-sdk.dev/docs/migration-guides/migration-guide-6-0
-
https://ai-sdk.dev/docs/troubleshooting/streaming-not-working-when-proxied
- https://github.com/vercel/ai/issues/10060
Closes AI-514
Related AI-509
## I have read the
[CONTRIBUTING.md](https://github.com/supabase/supabase/blob/master/CONTRIBUTING.md)
file.
YES
## What kind of change does this PR introduce?
- For the new filters experience, we are missing null changes for dates
and also no support for isNull
- Updated E2E tests to account for new null cases
`reasoningEffort: 'minimal'` was
[configured](https://github.com/supabase/supabase/blob/d5cc70560d/apps/studio/lib/ai/model.utils.ts#L55-L59)
in the provider registry but `getModel()` returns it as a separate value
that callers must destructure and forward — and 7 of 8 endpoints weren't
doing so. This meant `gpt-5-mini` (a reasoning model) was running at
default reasoning effort for every call.
This PR destructures `providerOptions` from `getModel()` and passes it
to `generateObject`/`generateText` in all affected endpoints.
## Benchmark (local, median of 5 runs)
| Endpoint | Before (s) | After (s) | Speedup |
|----------|-----------|----------|---------|
| title-v2 | 7.0 | 1.9 | 3.7x |
| cron-v2 | 2.3 | 0.9 | 2.6x |
| filter-v1 | 5.8 | 2.2 | 2.6x |
| feedback/classify | 3.5 | 0.9 | 3.9x |
| feedback/rate | 2.9 | 0.9 | 3.2x |
`code/complete` and `policy` also received the fix but aren't
benchmarked here as they require a live DB connection and use multi-step
tool calls (separate latency concern tracked in AI-419).
To test the SQL naming, visit the SQL Editor in sidebar, add some SQL
like:
```sql
create table todos (
id serial primary key,
task text not null,
completed boolean default false
);
```
Right click on the snippet, "Rename" and "Rename with Supabase AI"
Closes AI-443
Enables Braintrust tracing for AI Assistant chats to support debugging
and future online evals.
**Code Changes**
- Wraps `generateAssistantResponse` in a Braintrust `traced()` span,
logging the user's latest message as input along with metadata
(`chatId`, `chatName`, `projectRef`, `userId`, `orgId`, `planId`, etc.)
- Threads JWT claims from `apiWrapper` → handler to log `userId` in
Braintrust without an extra API call (+ expanded `apiWrapper` tests)
- Threads `orgId` and `planId` from `getOrgAIDetails` to log in
Braintrust
**Infrastructure Changes**
- Created a "Vercel" service account in Braintrust
- Added `BRAINTRUST_API_KEY` and `BRAINTRUST_PROJECT_ID` env vars to the
studio-staging project in Vercel using a service token for the above
service account
- Added an "Overview" view to the Logs tab in the Braintrust Assistant
project to surface the new metadata
**Precautions**
- HIPAA sensitive projects are excluded from logging (see
https://github.com/supabase/supabase/pull/42787 for the detection logic)
- Production is temporarily excluded from logging until we're confident
in the setup
**Testing steps**
- Chat with the AI Assistant in the [studio-staging preview
build](https://github.com/supabase/supabase/pull/42963#issuecomment-3917178023)
below
- Visit the [Logs tab in the Braintrust Assistant
project](https://www.braintrust.dev/app/supabase.io/p/Assistant/logs)
and inspect the trace
<img width="4680" height="962" alt="CleanShot 2026-02-18 at 17 43 55@2x"
src="https://github.com/user-attachments/assets/c3a11b21-4e7f-4e90-bdab-a25ab8ee0d1f"
/>
<img width="2632" height="1288" alt="CleanShot 2026-02-18 at 17 45
04@2x"
src="https://github.com/user-attachments/assets/6c7b6ebc-5090-4ede-8f71-859ff7e386aa"
/>
**References**
- https://www.braintrust.dev/docs/integrations/sdk-integrations/vercel
- https://www.braintrust.dev/docs/instrument/custom-tracing
Closes AI-438
Detects HIPAA customers server-side in the assistant code path. Threads
`isHipaaEnabled` boolean through `getOrgAIDetails` → `generate-v4` →
`generateAssistantResponse`. The motivation is to support online evals
down the road, where we'll want to exclude HIPAA projects from Assistant
tracing.
This PR follows existing patterns for checking if HIPAA is enabled for a
project (org has HIPAA addon + project is sensitive). Example
[[1]](https://github.com/supabase/supabase/blob/a5dd0a96716561443778f38a518b61d6cac95c19/apps/studio/components/interfaces/Settings/Addons/Addons.tsx#L75),
[[2]](https://github.com/supabase/supabase/blob/6858d4e18d9359d573fe3dff73bc4e5fa1cfe219/apps/studio/hooks/misc/useOrgOptedIntoAi.ts#L69).
```ts
const hasHipaaAddon = subscriptionHasHipaaAddon(subscription) && settings?.is_sensitive
```
(I call it `isHipaaEnabled` in this PR to avoid it being misunderstood
as just the org-level addon, rather it's a combo of that addon being
present AND high compliance being enabled on the project).
### Verification steps
<details><summary>Click to view the steps I followed to sanity check it
works with the local stack</summary>
Tested locally with `mise fullstack`:
1. Found my org's subscription ID:
```sh
docker exec platform-db-1 psql -U postgres -c "SELECT id, customer_id,
status FROM orb.subscriptions;"
```
2. Added HIPAA addon to it:
```sh
docker exec platform-db-1 psql -U postgres -c "
UPDATE orb.subscriptions
SET price_intervals = price_intervals || '[{\"price\": {\"unit_config\":
{\"unit_amount\": \"350.00\"}, \"external_price_id\":
\"addon_security_hipaa\", \"item\": {\"name\": \"HIPAA\"}}}]'::jsonb
WHERE id = '<subscription_id>';"
```
2. Toggled on High Compliance (Project Settings → General)
3. Added a temporary log after `getOrgAIDetails` in `generate-v4.ts`:
```ts
console.log('[HIPAA]', { isHipaaEnabled })
```
4. Sent a message in the AI Assistant → `isHipaaEnabled: true`
5. Toggled off High Compliance → resent → `isHipaaEnabled: false`
6. Removed addon from subscription, left project toggle on →
`isHipaaEnabled: false`
```sql
-- Find addon index:
SELECT ordinality - 1 as idx FROM orb.subscriptions,
jsonb_array_elements(price_intervals) WITH ORDINALITY AS elem(val,
ordinality)
WHERE id = '<subscription_id>'
AND val->'price'->>'external_price_id' = 'addon_security_hipaa';
-- Remove by index:
UPDATE orb.subscriptions SET price_intervals = price_intervals - <idx>
WHERE id = '<subscription_id>';
```
All three cases confirm `isHipaaEnabled` requires both the org addon and
the project-level toggle.
</details>
Closes AI-434
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **New Features**
* Added HIPAA mode detection and exposed it in AI workflows.
* API request functions now accept optional custom authorization headers
for downstream calls.
* **Tests**
* Added tests covering HIPAA scenarios and verifying authorization
header propagation in related flows.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
* handling and mock data support
* admonition
* only show in empty state
* rabbit
* mock without local overrides
* remove admonition
* make incident banner more prominent
* remove mock data
* move to tool
* prettier
* fix(studio): get_active_incidents tool
---------
Co-authored-by: Charis Lam <26616127+charislam@users.noreply.github.com>
* update onboarding
* update model and fix part issue
* action orientated assistant
* fix tool
* lock
* remove unused filter
* fix tests
* fix again
* update package
* update container
* fix tests
* ai realtime
* ai realtime
* refactor(ai assistant): break out message markdown and profile picture
* wip
* refactor(ai assistant): break up message component
* refactor: break ai assistant message down into multiple files
* add limitations prompt
* limitations prompt
* link prompt
* refactor: simplify ReportBlock state
* fix: styling of draggable report block header
When the drag handle is showing, it overlaps with the block header.
Decrease the opacity of the header so the handle can be seen and the two
can be distinguished.
* fix: minor tweaks to tool ui
* refactor: simplify DisplayBlockRenderer state
* fix: remove double deploy button in edge function block
When the confirm footer is shown, the deploy button on the top right should be
hidden (not just disabled) to avoid confusion.
* refactor, test: message sanitization by opt-in level
Refactor the message sanitization to have more type safety and be more testable.
Add tests to ensure:
- Message sanitization always runs on generate-v4
- Message sanitization correctly works by opt-in level
* Fix conflicts in pnpm lock
* Couple of nits and refactors
* Revert casing for report block snippet
* adjust sanitised prompt
* Fix tests
* empty states
* prompt otpimise
* refine prompt
* prompt optimizer
* remove realtime for now
* Update apps/studio/lib/ai/prompts.ts
Co-authored-by: Charis <26616127+charislam@users.noreply.github.com>
* Update apps/studio/lib/ai/prompts.ts
Co-authored-by: Charis <26616127+charislam@users.noreply.github.com>
* updates
* feature flag
* use flag and additional check
* remove sort
* messages copy
---------
Co-authored-by: Charis Lam <26616127+charislam@users.noreply.github.com>
Co-authored-by: Joshen Lim <joshenlimek@gmail.com>
* update onboarding
* update model and fix part issue
* action orientated assistant
* fix tool
* lock
* remove unused filter
* fix tests
* fix again
* update package
* update container
* fix tests
* refactor(ai assistant): break out message markdown and profile picture
* wip
* refactor(ai assistant): break up message component
* refactor: break ai assistant message down into multiple files
* add limitations prompt
* limitations prompt
* link prompt
* refactor: simplify ReportBlock state
* fix: styling of draggable report block header
When the drag handle is showing, it overlaps with the block header.
Decrease the opacity of the header so the handle can be seen and the two
can be distinguished.
* fix: minor tweaks to tool ui
* refactor: simplify DisplayBlockRenderer state
* fix: remove double deploy button in edge function block
When the confirm footer is shown, the deploy button on the top right should be
hidden (not just disabled) to avoid confusion.
* refactor, test: message sanitization by opt-in level
Refactor the message sanitization to have more type safety and be more testable.
Add tests to ensure:
- Message sanitization always runs on generate-v4
- Message sanitization correctly works by opt-in level
* Fix conflicts in pnpm lock
* Couple of nits and refactors
* Revert casing for report block snippet
* adjust sanitised prompt
* Fix tests
* prompt optimizer
* remove realtime for now
* Update apps/studio/lib/ai/prompts.ts
Co-authored-by: Charis <26616127+charislam@users.noreply.github.com>
* Update apps/studio/lib/ai/prompts.ts
Co-authored-by: Charis <26616127+charislam@users.noreply.github.com>
---------
Co-authored-by: Charis Lam <26616127+charislam@users.noreply.github.com>
Co-authored-by: Joshen Lim <joshenlimek@gmail.com>
* update onboarding
* update model and fix part issue
* action orientated assistant
* fix tool
* lock
* remove unused filter
* fix tests
* fix again
* update package
* update container
* fix tests
* refactor(ai assistant): break out message markdown and profile picture
* wip
* refactor(ai assistant): break up message component
* refactor: break ai assistant message down into multiple files
* refactor: simplify ReportBlock state
* fix: styling of draggable report block header
When the drag handle is showing, it overlaps with the block header.
Decrease the opacity of the header so the handle can be seen and the two
can be distinguished.
* fix: minor tweaks to tool ui
* refactor: simplify DisplayBlockRenderer state
* fix: remove double deploy button in edge function block
When the confirm footer is shown, the deploy button on the top right should be
hidden (not just disabled) to avoid confusion.
* refactor, test: message sanitization by opt-in level
Refactor the message sanitization to have more type safety and be more testable.
Add tests to ensure:
- Message sanitization always runs on generate-v4
- Message sanitization correctly works by opt-in level
* Fix conflicts in pnpm lock
* Couple of nits and refactors
* Revert casing for report block snippet
* adjust sanitised prompt
* Fix tests
---------
Co-authored-by: Charis Lam <26616127+charislam@users.noreply.github.com>
Co-authored-by: Joshen Lim <joshenlimek@gmail.com>
* try a really long context window to maximize caching
* update examples
* attempt to update packages and useChat
* update endpoints
* update zod
* zod
* update to v5
* message update
* Revert "zod"
This reverts commit ec39bac6b6.
* revert zod
* zod i
* fix complete endpoints
* remove async
* change to content
* type cleanup
* Revert the package bumps to rebuild them.
* Bump zod to 2.25.76 in all packages.
* Bump openai in all packages.
* Bump ai and ai-related packages.
* Remove unneeded files.
* Fix the rest of the migration stuff.
* Prettier fixes.
* testing new gptoss models
* sonnet as pro
* back to 3.7
* revert test
* add policy list tool
* fix
* refactor
* ai sdk 5 fixes
* refactor complete endpoint
* edge function prompt
* remove example
* use limited model for completions
* remove duplicate
* Update bedrock.ts
Co-authored-by: divit <27228526+delgado3d@users.noreply.github.com>
* add default check
* only add prefix if more then one region
* temp use api
* refactor to use openai api
* test
* remove unused import
* Prettier lint
* Fix test
* revert back to bedrock
* Tiny style fix for spacing of code blocks from edit message button
* try generate text normally
* prompt adjust
* add list style to prompt
* prompt adjust
* prompt adjust emphasise display query
* update prompts
* prompt adjust
* remove package
* prompt adjust
---------
Co-authored-by: Jordi Enric <jordi.err@gmail.com>
Co-authored-by: Ivan Vasilov <vasilov.ivan@gmail.com>
Co-authored-by: divit <27228526+delgado3d@users.noreply.github.com>
Co-authored-by: Joshen Lim <joshenlimek@gmail.com>
* get `stop()` from useChat and use it
* Implement abort signal in `generate-v4` API
* Enhance `onStop` functionality in AIAssistant to store partial message when aborted
* Update dependencies in AIAssistant to include `updateMessage` in effect hook (lint warning fix)
* change onStop prop to be required in AssistantChatForm
* add comment for saving partial AI responses
* remove redundant text size class
* Make onStop prop optional in AssistantChatForm and conditionally render stop button
* Add loading spinner to stop button in AssistantChatForm
* simplify opacity handling for submit button
* Add tooltips
* Improve UI error handler for assistant panel
* Small tweak
* Filter out any tool calls that are streaming in generate-v4, happens when aborting the assistant while a tool is being called
---------
Co-authored-by: Joshen Lim <joshenlimek@gmail.com>
* try a really long context window to maximize caching
* update examples
* attempt to update packages and useChat
* update endpoints
* update zod
* zod
* update to v5
* message update
* Revert "zod"
This reverts commit ec39bac6b6.
* revert zod
* zod i
* fix complete endpoints
* remove async
* change to content
* type cleanup
* Revert the package bumps to rebuild them.
* Bump zod to 2.25.76 in all packages.
* Bump openai in all packages.
* Bump ai and ai-related packages.
* Remove unneeded files.
* Fix the rest of the migration stuff.
* Prettier fixes.
* add policy list tool
* refactor
* ai sdk 5 fixes
* refactor complete endpoint
* edge function prompt
* remove example
* slight prompt change
* Minor clean up
* More clean up
---------
Co-authored-by: Jordi Enric <jordi.err@gmail.com>
Co-authored-by: Ivan Vasilov <vasilov.ivan@gmail.com>
Co-authored-by: Joshen Lim <joshenlimek@gmail.com>
* Clean up usage of newOrgAiOptIn and useBedrockAssistant feature flags
* Remove all OpenAI endpoints
* Fix for self-hosted
* Default isLimited to false
* Update PG meta tests
* Fix unit tests for model
* Revert pg meta tests
* Fix test
---------
Co-authored-by: Alaister Young <alaister@users.noreply.github.com>
* Validate the aiOptInLevel on the server
* Fix another issue when the project and mismatched org.
* Fix a type error.
---------
Co-authored-by: Ivan Vasilov <vasilov.ivan@gmail.com>
* wip
* feat: distribute bedrock requests across regions
* fix: tests and type errors
* chore: remove AWS_BEDROCK_REGION from turbo.json
* fix: vercel oidc provider to fallback in credential chain
---------
Co-authored-by: delgado3d <27228526+delgado3d@users.noreply.github.com>
Joshen Lim