HN 표시: Lightless Labs Refinery – 다중 모델 합의 및 합성

Iteratively reach consensus across multiple AI models cargo install --path crates/refinery_cli Requires Rust 1.85+. Add to your Cargo.toml [dependencies] refinery_core = "0.1" Set up credentials for at least one provider (see Credentials) refinery converge "What are the three most impactful breakthroughs in physics?" \ --models claude-code,codex-cli,gemini-cli Models propose, evaluate each other, and repeat until consensus. Pass models as a comma-separated list using provider/model format refinery converge "your prompt" --models claude-code/claude-opus-4-6,gemini-cli/gemini-3.1-pro-preview Short aliases use each provider's default model refinery converge "your prompt" --models claude-code,codex-cli,gemini-cli Refinery dispatches prompts to locally installed and authenticated CLI tools. Install and authenticate any of the supported CLIs, then pass them as --models : | Provider | Default model | CLI binary | Install | |---|---|---|---| claude-code | claude-opus-4-6 | claude | npm i -g @anthropic-ai/claude-code | codex-cli | gpt-5.4 | codex | npm i -g @openai/codex | gemini-cli | gemini-3.1-pro-preview | gemini | npm i -g @google/gemini-cli | opencode | (none — model required) | opencode | opencode.ai | Override the model with provider/model syntax: refinery converge "prompt" --models claude-code/claude-sonnet-4-6,codex-cli/o3-pro OpenCode supports multiple sub-providers. Use opencode/sub-provider/model : refinery converge "prompt" --models \ opencode/opencode/minimax-m2.5-free,\ opencode/kimi-for-coding/kimi-k2-thinking,\ opencode/minimax-coding-plan/MiniMax-M2.5,\ opencode/zai-coding-plan/glm-5 Run opencode models to list all available models. Use any combination of providers in a single run: refinery converge "prompt" --models \ claude-code,codex-cli,gemini-cli,\ opencode/kimi-for-coding/kimi-k2-thinking,\ opencode/zai-coding-plan/glm-5 Set the convergence threshold refinery converge "prompt" --models claude-code,codex-cli --threshold 9.0 Limit the number of rounds refinery converge "prompt" --models claude-code,codex-cli --max-rounds 3 Require more consecutive rounds of stable leadership before converging (must be between 1 and 20, and ) so models know which content came from where. Non-UTF-8 files and files exceeding the 1 MB budget are rejected with a clear error before any API calls are made. Pipe a prompt from another command (max 1 MB) cat question.txt | refinery converge - --models claude-code,codex-cli refinery converge "prompt" --models claude-code,codex-cli --verbose # per-round progress refinery converge "prompt" --models claude-code,codex-cli --debug # raw CLI invocations | Code | Meaning | |---|---| | 0 | Converged or single model | | 1 | Error or cancellation | | 2 | Max rounds exceeded | | 3 | Insufficient models | | 4 | Config or input error | The Hitchhiker's Guide to the Galaxy $ refinery converge --max-rounds 5 --output-format json --models claude-code,codex-cli,gemini-cli --timeout 1800 --idle-timeout 480 --output-dir out "What's the answer to life, the Universe, and everything?" Round 1/5 ── propose ── ✓ codex-cli/gpt-5.4 proposed (1 words) — "42." ✓ gemini-cli/gemini-3.1-pro-preview proposed (56 words) — "The answer to life, the Universe, and everything is **42**, ..." ✓ claude-code/claude-opus-4-6 proposed (72 words) — "42 — the answer computed by the supercomputer Deep Thought i..." ── evaluate ── ✓ gemini-cli/gemini-3.1-pro-preview → codex-cli/gpt-5.4: 8.0 — "The answer correctly identifies the iconic response to the p..." ✓ codex-cli/gpt-5.4 → gemini-cli/gemini-3.1-pro-preview: 9.0 — "This is a strong answer: accurate, clear, and appropriately ..." ✓ claude-code/claude-opus-4-6 → codex-cli/gpt-5.4: 8.0 — "The answer is correct and appropriately succinct. '42' is th..." ✓ gemini-cli/gemini-3.1-pro-preview → claude-code/claude-opus-4-6: 10.0 — "An excellent, comprehensive answer that not only provides th..." ✓ codex-cli/gpt-5.4 → claude-code/claude-opus-4-6: 9.0 — "This is a strong answer: accurate, well-written, and appropr..." ✓ claude-code/claude-opus-4-6 → gemini-cli/gemini-3.1-pro-preview: 8.0 — "This is a solid, accurate, and well-written answer. It corre..." → Not converged (9.5/8.0, stable 1/2) R1 claude-code/claude-opus-4-6 9.5 ★ gemini-cli/gemini-3.1-pro-preview 8.5 codex-cli/gpt-5.4 8.0 Round 2/5 ── propose ── ✓ codex-cli/gpt-5.4 proposed (11 words) — "42, according to Douglas Adams' *The Hitchhiker's Guide to t..." ✓ claude-code/claude-opus-4-6 proposed (100 words) — "**42** — the answer computed by the supercomputer Deep Thoug..." ✓ gemini-cli/gemini-3.1-pro-preview proposed (188 words) — "The answer to life, the Universe, and everything is **42**." ── evaluate ── ✓ claude-code/claude-opus-4-6 → codex-cli/gpt-5.4: 8.0 — "This is a correct, well-attributed, and concise answer. It i..." ✓ codex-cli/gpt-5.4 → claude-code/claude-opus-4-6: 9.0 — "This is a strong answer: accurate, well-written, and appropr..." ✓ codex-cli/gpt-5.4 → gemini-cli/gemini-3.1-pro-preview: 9.0 — "Th