- Home
- Skills
- Agents & Orchestration
- Multi-Model LLM Council
Works with the AI Agents you already use
Multi-Model LLM Council
Make Claude, ChatGPT, Gemini, Kimi, GLM, DeepSeek, Qwen, Manus and your local Ollama models solve a problem together: blind peer review with a ranked vote, or an open debate until they reach consensus.
$9
Multi-Model LLM Council
Example session with this skill installed
Run a council with qwen and medgemma (my local Ollama models), review method: A bat and a ball cost 1.10 in total. The bat costs 1.00 more than the ball. How much does the ball cost? Answer briefly.
- Read your context and instructions
- Compiled the multi-model llm council
Method: independent answers, then blind peer review and ranking. Members: qwen (Ollama CLI), medgemma (Ollama API). Saved to councils/<date>-a-bat-and-a-ball/report.md.
Council answer (written by qwen, the top-ranked member)
Final answer
The ball costs 0.05.
Where the council agreed
Both members agreed on the final answer (0.05) and the algebra behind it. Both identified that the intuitive answer of 0.10 fails the condition: if the ball cost 0.10, the bat would cost 1.10 and the total would be 1.20, not 1.10. Both reported high confidence.
Where it disagreed, and how it was resolved
The disagreement was about presentation, not the mathematics.
- Response A gave a short derivation with one variable.
- Response B gave a long step-by-step breakdown with assumptions, two variables and a check step.
The peer vote ranked Response A higher because it followed the request to answer briefly. Both answers are correct.
Remaining uncertainty
None for the mathematics: the problem has a unique solution, assuming normal two-decimal currency.
Peer ranking (Borda count)
| Rank | Member | Points |
|---|---|---|
| 1 | qwen | 2 |
| 2 | medgemma | 0 |
Members' final answers
qwen (Response A): Let b be the cost of the ball. The bat costs b + 1.00. Total: b + (b + 1.00) = 1.10, so 2b = 0.10 and b = 0.05. CONFIDENCE: high
medgemma (Response B): A step-by-step breakdown (assumptions, variables x and y, two equations, substitution, check) reaching the same answer: the ball costs 0.05. CONFIDENCE: high
Next steps
- Add a cloud model (Claude, ChatGPT or Gemini) for a stronger council:
council.py add claude --preset claude --via cli - Try the open debate method on a design question:
council.py ask "..." --method consensus --rounds 3
Connects securely to your tools. The creator never sees your data.
About this skill
One problem, many models, one answer you can trust more
Ask a single model and you get a single point of view, with its blind spots. This skill turns the models you already have into a council that works on the same problem and checks each other's work.
You choose everything
- Which models sit on the council. Nothing is preset. Claude, ChatGPT/GPT, Gemini, Kimi (Moonshot), GLM (Z.ai), DeepSeek, Qwen, Mistral, Manus, OpenRouter, local Ollama and LM Studio models, opencode, or any other CLI or OpenAI-compatible server.
- How each model is reached: through its API (your own key, read from an environment variable) or its CLI on your PC, using the login or subscription you already have (
claude,codex,gemini,ollama,opencode...). - The method, for every run:
| Method | How it works | Best for |
|---|---|---|
review | Each model answers alone, then reviews the others' answers blind (anonymous, shuffled), revises, and ranks them (Borda count, no voting for yourself). The top-ranked model, or a chair you name, writes the synthesis. | Checking work, finding errors, avoiding groupthink |
consensus | The models debate openly, in turn, editing one shared draft until every model accepts it, or the rounds run out. Holdouts are named. | Design, planning, open-ended problems |
What you get
- A final answer that says where the models agreed, where they disagreed, and what is still uncertain.
- A readable
report.mdand a fulltranscript.jsonof every round, so you can audit the reasoning. - Claude in your current session can sit on the council too.
Built to run unattended
- Python standard library only: nothing to
pip install. - A member that fails (bad key, quota, timeout) is reported and dropped; the council carries on.
doctor --pingchecks that every member really answers before you spend money on a council.- API keys never go into files: only the name of the environment variable is stored.
Honest limits
Models can agree and still be wrong, and models trained on similar data share blind spots. The skill reports real dissent instead of smoothing it over, and tells you to check facts, numbers and citations.
One-time purchase: install and use it in your own projects and your own team, with free updates when a new version is published.
Requires Python 3.9+ and at least 2 models: your own API keys and/or model CLIs installed and logged in on your PC. API calls cost money and CLI calls count against subscription limits; each extra round adds one call per model. Manus runs a full agent task per call, so it is slow and uses Manus credits. Models can agree and still be wrong: check facts, numbers and citations.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 6 days ago
Requires
Permissions
Allowed Hosts
File Scopes
Terminal: runs the bundled Python script and the model CLIs you choose. Network: calls the APIs of the models you add. Environment variables: reads your API keys (only the variable names are stored, never the keys). Write files: saves council.json and the reports in councils/. The services listed are examples; you connect only the models you choose.