Make the call you
won't regret.
Council of Hats gives Hermes five decision frameworks and a deterministic question engine — so high-stakes calls get interrogated, logged with their rationale, and reviewed against how they actually turned out.
curl -fsSL antharmaya.com/council-of-hats/install.sh | bashAgents are confident. Confidence isn't a decision process.
Left alone, an agent picks the first plausible option, never names the trade-off, never records why, and never checks whether it was right. Council of Hats turns that into a discipline — without slowing anything down.
Five hats. One sharper decision.
Trade-Off Matrix
Systems trade-off analysis
What are we really comparing, what constraint dominates, and what are we quietly deprioritizing?
Failure Modes
Distributed failure analysis
What's the blast radius? Which failures are we designing for — and which are we choosing to ignore?
End-to-End Verification
End-to-end integrity
Where could silent corruption hide, and what actually verifies correctness across the boundary?
Ethics Triage
Privacy & ethics review
What user data does this touch, what's the consent mechanism, and what's the retention policy?
Agent Trajectory
Agent trajectory evaluation
Was the reasoning sound, was confidence calibrated, and was the final decision actually correct?
Detect. Question. Log. Review.
Detect
Hermes spots decision-relevant context and loads the matching pattern — no prompting required.
Question
A deterministic engine asks 3–5 structured questions. No LLM, no stochastic drift, stdlib only.
Log
Answers, rationale, confidence and framework are written to a local SQLite decision log.
Review
Revisit past decisions, mark outcomes, and stop repeating the calls that already went wrong.
$ hermes tool ask_structured_question --pattern tradeoff_matrixDeterministic. Local. Yours.
The question engine is pure Python stdlib — deterministic, instant, and free. No tokens, no API key.
Every decision lives in ~/.hermes/decisions/decision_log.db. Yours, on your machine, forever.
Trade-offs, failure modes, end-to-end integrity, ethics, and agent trajectory — extend with your own.
A built-in checklist fires before any feature that touches user data. Consent and retention, every time.
Council decides. Memory Bridge remembers — and checks if it worked.
Each tool stands alone. Run together, they close the loop: a decision made through a Council pattern is consolidated by Memory Bridge, tracked across every agent, and surfaced again the next time the same question comes up.
Give your agents a decision process.
curl -fsSL antharmaya.com/council-of-hats/install.sh | bashMIT licensed · Python 3.11+ · Hermes 0.16+ · No API keys