Open
Conversation
Multi-provider (Anthropic, OpenAI, Gemini) endpoint for extracting discourse graph nodes from uploaded PDFs.
|
This pull request has been ignored for the connected project Preview Branches by Supabase. |
Widen Message.content to support multimodal content blocks and add systemPrompt/responseMimeType to Settings. Each provider's formatRequestBody now handles both text-only chat and PDF extraction, eliminating the parallel PROVIDERS block in the extraction route. OpenAI extraction switches from Responses API to Chat Completions (now supports PDF). Gemini field casing fixed to match REST API docs.
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
Add structured output enforcement via each provider's native mechanism: Anthropic output_config, OpenAI response_format with strict mode, Gemini responseJsonSchema. Removes prompt-based JSON instructions and response cleanup parsing since constrained decoding guarantees valid JSON.
Per AGENTS.md: functions with more than 2 parameters use named parameters via object destructuring.
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
Collaborator
Author
|
@coderabbitai full review |
Contributor
✅ Actions performedFull review triggered. |
Inline disables for response_format, json_schema (OpenAI), and output_config (Anthropic) — external API contract names.
Contributor
📝 WalkthroughWalkthroughThis PR introduces a PDF-to-AI extraction feature with a new POST API endpoint that accepts PDF data, selects LLM providers (Anthropic, OpenAI, Gemini), and extracts structured discourse graph nodes. It includes type definitions, prompt templates, response parsing logic, and updates to provider implementations to support system prompts and JSON output schemas. Changes
Sequence DiagramsequenceDiagram
participant Client
participant APIRoute as API Route<br/>(extract/route.ts)
participant ProviderConfig as Provider Config<br/>(providers.ts)
participant LLMProvider as LLM Provider<br/>(Anthropic/OpenAI/Gemini)
participant ResponseParser as Parser<br/>(parseExtractionResponse)
Client->>APIRoute: POST with PDF, provider, model
APIRoute->>APIRoute: Validate request<br/>against schema
APIRoute->>APIRoute: Read provider API key<br/>from env
APIRoute->>ProviderConfig: Build messages &<br/>settings with<br/>systemPrompt/outputSchema
ProviderConfig->>ProviderConfig: Format provider-specific<br/>request payload
ProviderConfig-->>APIRoute: Return formatted<br/>request config
APIRoute->>LLMProvider: POST with timeout<br/>signal
LLMProvider-->>APIRoute: Response text<br/>(or error)
APIRoute->>ResponseParser: Parse extracted<br/>content JSON
ResponseParser->>ResponseParser: Validate against<br/>ExtractionResultSchema
ResponseParser-->>APIRoute: Typed ExtractionResult
APIRoute-->>Client: { success, data }<br/>or { success, error }
Estimated code review effort🎯 4 (Complex) | ⏱️ ~45 minutes Possibly related PRs
🚥 Pre-merge checks | ✅ 3✅ Passed checks (3 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
🧹 Nitpick comments (2)
apps/website/app/utils/llm/providers.ts (1)
20-29: Add eslint-disable comments for API-required snake_case properties.The
response_formatandjson_schemaproperties are mandated by the OpenAI API. Consider adding inline eslint-disable comments to silence the warnings and document why.🔧 Proposed fix
...(settings.outputSchema && { + // eslint-disable-next-line `@typescript-eslint/naming-convention` response_format: { type: "json_schema", + // eslint-disable-next-line `@typescript-eslint/naming-convention` json_schema: { name: "extraction_result", strict: true, schema: settings.outputSchema, }, }, }),🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@apps/website/app/utils/llm/providers.ts` around lines 20 - 29, The OpenAI API requires snake_case keys `response_format` and `json_schema` inside the object created when `settings.outputSchema` is present; add inline eslint-disable comments (e.g., // eslint-disable-next-line `@typescript-eslint/naming-convention`) immediately above or inline with those properties to suppress naming-convention errors and include a short comment noting “required by OpenAI API” to document why the rule is disabled; update the object around `settings.outputSchema`, `response_format`, `json_schema`, and the `name: "extraction_result"` entry accordingly.apps/website/app/types/extraction.ts (1)
35-55: Consider generating JSON schema from Zod to prevent drift.The
EXTRACTION_RESULT_JSON_SCHEMAmanually mirrorsExtractionResultSchema. If these diverge, the LLM's structured output may not match runtime validation. Libraries likezod-to-json-schemacan generate one from the other.#!/bin/bash # Check if zod-to-json-schema is already a dependency rg -l "zod-to-json-schema" apps/website/package.json🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@apps/website/app/types/extraction.ts` around lines 35 - 55, The JSON schema constant EXTRACTION_RESULT_JSON_SCHEMA is manually duplicated and can drift from the Zod schema (ExtractionResultSchema); replace the hand-written EXTRACTION_RESULT_JSON_SCHEMA with a generated schema by using zod-to-json-schema (or equivalent) to convert ExtractionResultSchema into JSON Schema at build/runtime, update imports so ExtractionResultSchema is the sole source of truth, and export the generated schema where EXTRACTION_RESULT_JSON_SCHEMA was used so consumers (LLM structured output validation) always get the schema derived from the Zod definition.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Nitpick comments:
In `@apps/website/app/types/extraction.ts`:
- Around line 35-55: The JSON schema constant EXTRACTION_RESULT_JSON_SCHEMA is
manually duplicated and can drift from the Zod schema (ExtractionResultSchema);
replace the hand-written EXTRACTION_RESULT_JSON_SCHEMA with a generated schema
by using zod-to-json-schema (or equivalent) to convert ExtractionResultSchema
into JSON Schema at build/runtime, update imports so ExtractionResultSchema is
the sole source of truth, and export the generated schema where
EXTRACTION_RESULT_JSON_SCHEMA was used so consumers (LLM structured output
validation) always get the schema derived from the Zod definition.
In `@apps/website/app/utils/llm/providers.ts`:
- Around line 20-29: The OpenAI API requires snake_case keys `response_format`
and `json_schema` inside the object created when `settings.outputSchema` is
present; add inline eslint-disable comments (e.g., // eslint-disable-next-line
`@typescript-eslint/naming-convention`) immediately above or inline with those
properties to suppress naming-convention errors and include a short comment
noting “required by OpenAI API” to document why the rule is disabled; update the
object around `settings.outputSchema`, `response_format`, `json_schema`, and the
`name: "extraction_result"` entry accordingly.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Repository UI
Review profile: CHILL
Plan: Pro
Run ID: 0eb45979-82e8-48e9-a803-029f780a6309
📒 Files selected for processing (6)
apps/website/app/api/ai/extract/route.tsapps/website/app/prompts/extraction.tsapps/website/app/types/extraction.tsapps/website/app/types/llm.tsapps/website/app/utils/ai/parseExtractionResponse.tsapps/website/app/utils/llm/providers.ts
Gemini parts use { text } not { type: "text", text }. The shared
textBlock was using the Anthropic/OpenAI format which Gemini rejects.
sid597
changed the title
mdroidian
approved these changes
Apr 3, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
https://www.loom.com/share/b6fc57a6040c41dabf4155f61bcb2df0
Summary by CodeRabbit
Release Notes
New Features