Provider interoperability
This workflow friction appears across 11 independent open-source repositories. Repetition raises confidence, but direct user validation still decides whether it is worth building.
Where this need appears
Each item below is a public, qualified GitHub discussion. The Radar groups related friction, while the original Issue remains available for source checking.
- Add gemini-3.1-pro-previewgoogle-gemini/gemini-cli · 127 comments · 110 positive reactions
- The batch method from ChatModels and all the Runnables does not really support the OpenAI batch API.langchain-ai/langchain · 17 comments · 37 positive reactions
- [Feature]: Sign in with ChatGPT (Codex subscription) as an LLM providerTHU-MAIC/OpenMAIC · 9 comments · 3 positive reactions
- Temporal Knowledge Leakage in LLM Backtesting (Model-Level Future Information Contamination)TauricResearch/TradingAgents · 5 comments · 3 positive reactions
- feat: Add support for OpenAI-compatible LLM endpoints (Groq, Ollama, Azure OpenAI, etc.)google/adk-java · 11 comments · 3 positive reactions
- feat: 本地运行模式、Gemini/Codex 适配器、i18n 国际化、定时启动及项目管理增强oritera/Cairn · 1 comments · 7 positive reactions
- [BUG] Native gemini provider: 400 "Unknown name additionalProperties" in tool schemas (regression vs 0.3.1)afjcjsbx/picoclaw · 8 comments · 0 positive reactions
- Provider support for custom/local OpenAI-compat endpoints (Codex + OpenCode)nanocoai/nanoclaw · 6 comments · 0 positive reactions
Turn the signal into one observable action
Deliver the outcome manually or with a narrow prototype before building a reusable feature.
A user completes the real workflow and commits time, data, distribution or budget to repeat it.
Ask before you build
- When did you last need this?
- What outcome were you trying to reach?
- What did you use instead?
- How often does this occur?
- What commitment would prove it matters?
Professional guardrail: Use a fixed test case and compare completion, regressions and recovery against the unchanged baseline.
Choose a narrow next move
Open the public guided brief, interview someone who recently hit it, and record one concrete commitment or decision in your browser. Do not treat issue counts or compliments as willingness to pay.