🐙 Eight tentacles, zero blind spots.
Claude Octopus now orchestrates 8 AI providers — one per tentacle:
| Tentacle | Provider | Cost |
|---|---|---|
| 🔴 | Codex (OpenAI) | OAuth (free with ChatGPT sub) |
| 🟡 | Gemini (Google) | OAuth (free with Google AI sub) |
| 🟣 | Perplexity | API key |
| 🌐 | OpenRouter | API key (100+ models) |
| 🟢 | Copilot (GitHub) | Free with Copilot subscription |
| 🟠 | Qwen (Alibaba) | FREE — 1,000-2,000 requests/day [NEW] |
| ⚫ | Ollama (Local) | Free (runs locally) |
| 🔵 | Claude (Anthropic) | Included with Claude Code |
New in v9.10.0
Qwen CLI as 8th provider — Fork of Gemini CLI with free OAuth tier. Agent types: qwen, qwen-research. Install: npm install -g @qwen-code/qwen-code, then run qwen to authenticate via OAuth.
Hardened OpenRouter (P1) — Added --max-time 60 timeout, HTTP status code handling, 429 retry with Retry-After header, deduplicated execute functions.
Robust json_extract (P0) — Replaced brittle regex with 3-tier fallback: jq → python3 → improved regex. Old regex broke on whitespace, escaped quotes, arrays.
Copilot Coding Agent files — .github/agents/*.agent.md for all 10 native agents. GitHub's server-side coding agent now discovers and uses our agents.
Gemini .toml custom commands — .gemini/commands/octo/ with 4 persona commands for interactive human use.
Gemini test suite — 44 new tests covering all Gemini integration points.
DeepSeek model update — deepseek-r1 → deepseek-r1-0528.
Testing
135/135 tests passing across all suites (adapter-flags: 23, openclaw: 31, copilot: 37, gemini: 44).
Full Changelog: v9.9.3...v9.10.0