TITAN v7.1.0 "Council" 🧠

Local-first intelligence, together — agents advising agents, models sized to their machines, and benchmarks you can trust.

Highlights

🧠 /moa — Mixture of Agents: ask a council, not a model. N advisors answer in parallel (your own local models, across your own machines — $0/turn), one aggregator synthesizes with full tool use. Failures footnote, slow advisors drop, the turn never aborts.
🌐 The MCP triangle: TITAN is now an agent-level MCP tool (titan_chat, titan_delegate_task, titan_moa, titan_status) inside Hermes Agent and OpenClaw — and consumes their MCP servers right back. One config line each. See docs/INTEROP.md.
📏 Context-Fit: TITAN learns each deployment's real context ceiling from live traffic and sizes its toolset to fit — small-context local deployments degrade gracefully instead of failing, and every request got ~4K tokens lighter.
🧪 TITAN_BENCH=1 isolation mode + corrected benchmarks: Qwen3.6-35B-A3B NVFP4 on DGX-Spark-class hardware scores 85% through TITAN's harness — tying the RTX 5090's best (benchmarks/MODEL_COMPARISON.md).
vitest 4 migration, full dependency sweep (incl. ws security fix), Welcome-Mode onboarded fix.

Suite: 372 files / 8,137 passing / 0 failing. Upgrade: npm install -g titan-agent@latest (migrations are automatic; docs/MIGRATION.md).