← back to the overview
In practice

Fifteen ways
this earns its keep.

The overview says what a council is. This is what it looks like on a Tuesday — the material you hand it, where it sits in your work, how far you turn it up, and the machinery that keeps the result honest.

What you hand it

Anything you would hand a colleague.

chapter-12.md criteria.md VERDICT: FIX THESE FIRST Specs, plans, chapters — any material plus criteria "council review this chapter against my style guide"

Verdict on Anything

Not just code — hand the council a spec, a plan, even a book chapter, plus the criteria that matter to you. Reviewers judge it independently and you get one adjudicated verdict: what to fix first. Flip the ask and the same bench writes the chapter instead.

schema-plan.md gpt $0.09 gemini $0.14 deepseek $0.06 glm $0.04 llama3 · LOCAL $0.00 ↑ the local seat caught it a member, not a mode Local models take an equal seat — private and $0 amicus council run --prompt-file schema-plan.md --models gemini,gpt,llama3

Same Table

Seat a local model at the same table as the cloud ones — same prompt, same vote, $0.00, and nothing leaves your machine. Sometimes the free seat is the one that catches it.

The bench producing a deliverable: three seats each write a response, judges rank which best does the work, the chair closes with an ANSWER THE BENCH DOES THE WORK brief.md what to produce gemini draft A gpt draft B deepseek draft C which best does the work? 1st draft C 2nd draft A names stripped chair · not Claude ANSWER: Converged merged · residual risk named

Hand Over the Work

Not every ask is “check this.” Point the same bench at the job itself and every seat produces the deliverable — the analysis, the plan, the draft. The judges rank which response best does the work, and the chair merges the strongest into one answer, naming the claims it still rests on.

Mon · council run Fix these first rewrite Today · continue Ship it $ amicus list 7f3a · mon 81c2 · wed 9d40 · today same session, new evidence

Pick Up the Thread

Monday's council flagged two problems, so you rewrote. Pick up the same thread today — one question, and yesterday's findings flip from red to green.

Where it fits

In the editor, in the pipeline, in front of you.

release pipeline build ✓ tests ✓ council … diff · 4,012 lines amicus council run gemini gpt deepseek reviews → cross-review → tally → chair verdict.json exit 0 VERDICT: Ship it exit 1 Fix these first

Ship Gate

Release night, a 4,000-line RC diff nobody fully read. Put a council between your pipeline and the tag — exit 0 ships it, exit 1 sends it back.

Claude Code mcp servers db github repomix auto-discover amicus council gemini gpt llama3 · local db github repomix Every MCP server rides along into every seat auto-discovered — zero extra setup

Tools Travel

Configure an MCP server once in Claude Code and every seat picks it up — Gemini, GPT, even your local model. Your tools travel with the council, zero extra setup.

$ amicus models --check --live alias provider status gemini google serves ✓ gpt openai serves ✓ deepseek openrouter serves ✓ gpt-pro openai not serving · 404 caught before the council paid for it glm openrouter serves ✓ Probe the live catalog before you gate on it amicus models --check --live

Prove It Serves

Model catalogs rot quietly — an alias that worked last week can 404 today. One live probe proves every seat on your bench actually serves, so you find out in the check, not mid-council.

Council Workspace reviews ✓ · cross-review ● · tally ○ · chair ○ phase 2/4 gemini judged $0.04 gpt reviewing… judged $0.06 deepseek judged $0.03 glm judged $0.05 $0.07 qwen reviewing… $0.02 matrix cost by seat live for running councils · replayable for finished ones

Watch It Think

Open the Workspace and watch your council think — seats reporting in, phases advancing, costs ticking per model. Finished runs replay end to end.

How far you take it

Every dial is off until you name it.

DIRECT your API keys OPENROUTER 200+ models · one key LOCAL ollama · lm studio · no key gemini-pro long context gpt reasoning glm $0.04/run qwen code specialist llama3 $0 · private two and up — any bench size different families, different blind spots

Build the Bench

Models from the same family share the same blind spots — so build a bench that mixes families. Seat as many as you like — two and up — from your own API keys, one OpenRouter key, or local runtimes, each pick earning its chair on specialty, price, or privacy.

amicus council run reviews → cross-review → tally → chair the ritual — always the same all default off — nothing turns on unless you name it critic seat one reviewer turns adversarial expert lenses business · technical · customer · financial debate round contested findings defend, amend, or withdraw claude in the council judged by the bench — never votes, never chairs run shape 3 models × 2 waves + 1 chair critic seat + debate ON ~7 base runs + up to 6 rebuttal calls proceed?

Dial It Up

The ritual never changes — critic seat, expert lenses, debate rounds, and a Claude seat are all off until you name them. Dial up what you need and see the exact run shape before anything launches.

queue max-cost $2 launched 3 reviewers + chair ~7 calls · est $0.42 3 reviewers + chair ~7 calls · est $0.42 (o3-pro bench) est $31.80 BUDGET GATE est $0.42 charged $0.42 Cost shown before launch — runaway runs never start amicus council run --prompt-file plan.md --max-cost 2

Priced Up Front

Every council prints its estimated cost before a single call goes out — and a hard budget cap bounces the run you didn't mean to launch. The receipt matches the estimate, not your worst fear.

Why the answer holds

The parts that keep a council from being a poll.

Gemini GPT DeepSeek ANONYMIZE REVIEW A 2nd REVIEW B 1st REVIEW C 3rd disputed its own Reviews are ranked blind — self-bias washes out amicus council run --prompt-file design.md --models gemini,gpt,deepseek

Blind Ballots

Your council ranks every review blind — bylines stripped before the vote, so no model can favor its own. Some have even disputed their own work.

claude orchestrates votes gpt CHAIR gemini deepseek local VERDICT: Fix these first Claude orchestrates — a non-Claude chair signs the verdict amicus council run --prompt-file plan.md --chair deepseek

The Recusal

Your plan came from Claude — don't let Claude grade its own homework. A non-Claude chair runs the vote and signs the verdict; Claude only keeps the room in order.

council · adjudication Review A Review B Review C two agreed. one saw it. auth-refactor.diff session fixation on token refresh blocker CONFIRMED Unanimous councils can be wrong — disagreement is the signal amicus council run --prompt-file auth-refactor.diff --debate

Dissent Is Signal

Two reviewers pass your auth refactor; the third catches session fixation. Amicus treats that dissent as the finding — adjudicated and confirmed, not averaged away.

amicus workspace gemini $0.14 gpt-5.5 $0.11 deepseek $0.09 DEAD qwen $0.12 cli $ amicus council run ! seat lost: deepseek — exit 2 verdict.json "degraded": true ! degraded report.html

Never Silent

A seat dies mid-run? Every surface tells you — the CLI exits 2, verdict.json flips "degraded": true, and the report wears a red banner. Amicus never degrades silently.

Install it →