starred/karakeep

Fork 0

mirror of https://github.com/karakeep-app/karakeep.git synced 2026-04-25 07:56:05 +03:00

[PR #1907] [CLOSED] feat: Replace Chat Completions with Responses API for all OpenAI models #1924

New issue

Closed

opened 2026-03-02 11:59:47 +03:00 by kerem · 0 comments

kerem commented

2026-03-02 11:59:47 +03:00

Owner

📋 Pull Request Information

Original PR: https://github.com/karakeep-app/karakeep/pull/1907
Author: @camjac251
Created: 9/7/2025
Status: ❌ Closed

Base: main ← Head: feat/openai-responses-api

📝 Commits (1)

01a5a62 feat: Replace Chat Completions with Responses API for all OpenAI models

📊 Changes

4 files changed (+171 additions, -139 deletions)

View changed files

📝 packages/shared/config.ts (+9 -2)
📝 packages/shared/inference.ts (+148 -58)
📝 packages/shared/package.json (+1 -1)
📝 pnpm-lock.yaml (+13 -78)

📄 Description

Summary

Complete migration from Chat Completions API to Responses API
Add support for GPT-5 features (reasoning effort, verbosity control)
Model-aware parameter handling (GPT-4 vs GPT-5 capabilities)
Upgrade OpenAI SDK from v4.104.0 to v5.19.1
Update default models to GPT-5-mini for future readiness
No changes to InferenceClient public interface

New Configuration Options

INFERENCE_REASONING_EFFORT: Control reasoning depth (minimal/low/medium/high, default: low)
INFERENCE_VERBOSITY: Output verbosity for GPT-5 models (low/medium/high, default: medium)
Default models updated to gpt-5-mini (configurable via INFERENCE_TEXT_MODEL and INFERENCE_IMAGE_MODEL)

Why this change?

OpenAI's Responses API is the future-facing interface for all their models. Chat Completions is being phased out. This migration is necessary for GPT-5 compatibility and new features like reasoning control.

Key Implementation Details

GPT-5 models receive verbosity and full reasoning effort parameters
GPT-4 models get empty reasoning object and max_output_tokens parameter
Image inference restructured to use input_image format
Response extraction handles multiple API response structures
Structured output uses direct Zod schema conversion
Removed dependency on zodResponseFormat helper

Known Limitations

The OpenAI SDK v5 doesn't export TypeScript types for the Responses API yet, requiring use of any types in 4 places (with eslint-disable comments)
Once OpenAI adds proper type definitions, these can be replaced with proper types

Testing

pnpm typecheck --filter=@karakeep/shared --filter=@karakeep/trpc
pnpm lint --filter=@karakeep/shared --filter=@karakeep/trpc
Manual testing with real API keys:
- ✅ GPT-4 and GPT-5 model compatibility verified
- ✅ Text inference (bookmark summaries, general queries)
- ✅ Structured output (auto-tagging with Zod schemas)
- ✅ Image analysis (screenshot processing, base64 images)
- ✅ All inference modes tested with production-like payloads

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/karakeep-app/karakeep/pull/1907 **Author:** [@camjac251](https://github.com/camjac251) **Created:** 9/7/2025 **Status:** ❌ Closed **Base:** `main` ← **Head:** `feat/openai-responses-api` --- ### 📝 Commits (1) - [`01a5a62`](https://github.com/karakeep-app/karakeep/commit/01a5a629cf110c5e3e859f2d67df3335918288f7) feat: Replace Chat Completions with Responses API for all OpenAI models ### 📊 Changes **4 files changed** (+171 additions, -139 deletions) <details> <summary>View changed files</summary> 📝 `packages/shared/config.ts` (+9 -2) 📝 `packages/shared/inference.ts` (+148 -58) 📝 `packages/shared/package.json` (+1 -1) 📝 `pnpm-lock.yaml` (+13 -78) </details> ### 📄 Description ## Summary - **Complete migration** from Chat Completions API to Responses API - Add support for GPT-5 features (reasoning effort, verbosity control) - Model-aware parameter handling (GPT-4 vs GPT-5 capabilities) - Upgrade OpenAI SDK from v4.104.0 to v5.19.1 - Update default models to GPT-5-mini for future readiness - No changes to InferenceClient public interface ## New Configuration Options - `INFERENCE_REASONING_EFFORT`: Control reasoning depth (`minimal`/`low`/`medium`/`high`, default: `low`) - `INFERENCE_VERBOSITY`: Output verbosity for GPT-5 models (`low`/`medium`/`high`, default: `medium`) - Default models updated to `gpt-5-mini` (configurable via `INFERENCE_TEXT_MODEL` and `INFERENCE_IMAGE_MODEL`) ## Why this change? OpenAI's Responses API is the future-facing interface for all their models. Chat Completions is being phased out. This migration is necessary for GPT-5 compatibility and new features like reasoning control. ## Key Implementation Details - GPT-5 models receive verbosity and full reasoning effort parameters - GPT-4 models get empty reasoning object and `max_output_tokens` parameter - Image inference restructured to use `input_image` format - Response extraction handles multiple API response structures - Structured output uses direct Zod schema conversion - Removed dependency on `zodResponseFormat` helper ## Known Limitations - The OpenAI SDK v5 doesn't export TypeScript types for the Responses API yet, requiring use of `any` types in 4 places (with eslint-disable comments) - Once OpenAI adds proper type definitions, these can be replaced with proper types ## Testing - `pnpm typecheck --filter=@karakeep/shared --filter=@karakeep/trpc` - `pnpm lint --filter=@karakeep/shared --filter=@karakeep/trpc` - Manual testing with real API keys: - ✅ GPT-4 and GPT-5 model compatibility verified - ✅ Text inference (bookmark summaries, general queries) - ✅ Structured output (auto-tagging with Zod schemas) - ✅ Image analysis (screenshot processing, base64 images) - ✅ All inference modes tested with production-like payloads --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

kerem

2026-03-02 11:59:47 +03:00

closed this issue
added the
pull-request
label

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

starred/karakeep#1924

No description provided.

Rows
Columns