E3D Pilot

Product

The multi-model debate.

One tool call starts a debate. You get back a conclusion, the reasoning behind it, and a record of where the models disagreed.

Create an account Connect your editor

How it works

How a debate runs

  1. You ask

    Your editor sends the question and any context you choose to include.

  2. Models answer

    Several models respond independently.

  3. Models review

    Each model critiques the others and revises its position.

  4. You receive a result

    A single conclusion, the supporting arguments, and the open disagreements.

What comes back

What a debate returns.

A finished debate comes back in four parts. The numbers on the example match the list.

  1. Final answer

    The position the models converged on, in plain language.

  2. Agreement and disagreements

    How far the models agree, and every point they could not settle, so you can decide.

  3. What each model said

    Each model’s own position, labelled with the model that took it.

  4. Credits and record

    The credits the debate used, how long it is kept, and the full transcript to read or download from History.

A debate is a summary of what AI models argued, not advice. Models can be wrong, and agreement does not make an answer correct.

Question

Should the billing service move session storage from Redis to Postgres?

Final answer

Move to Postgres with a TTL index, after a read-latency check.

Agreement

Most of the models agree. 1 point of disagreement Logout fan-out may still need a cache.

What each model said

  • Model A

    Yes. One datastore fewer; the write volume is small.

  • Model B

    Yes, with a TTL index for expiry.

  • Model C

    Not yet. Measure read latency first.

Credits used · Saved until a set date · Full transcript

Illustrative example of a finished debate. Model names and text are made up.

What you get

A second, third and fourth opinion, on the record.

  • Independent answers

    Each model answers on its own before it sees anyone else’s work, so one model’s blind spot does not become everyone’s.

  • Cross-review

    The models critique each other and revise, so weak claims are challenged before they reach you.

  • A record in History

    Every debate is saved to your account with its answer, disagreements and transcript. Delete any debate whenever you like.

  • Spending you control

    Prepaid credits in one wallet per account. API keys are optional, and each key can have its own spending cap.

Connect

Connect your editor

Point your client at https://pilot.e3d.ai/mcp with an API key from your dashboard. Setup guides for each client are in the documentation.

These are the clients we have tested. Each connects with an API key; you can also ask from the web without one.

Copilot in VS Code

Add the server to .vscode/mcp.json or your user mcp.json. VS Code asks for the key once and keeps it out of the file.

{
  "servers": {
    "pilot": {
      "type": "http",
      "url": "https://pilot.e3d.ai/mcp",
      "headers": {
        "Authorization": "Bearer ${input:pilot-key}"
      }
    }
  },
  "inputs": [
    {
      "type": "promptString",
      "id": "pilot-key",
      "description": "E3D Pilot API key",
      "password": true
    }
  ]
}

Claude Code

Debates take minutes, so raise the tool timeout, then add the server with the key in a header.

export PILOT_KEY=YOUR_API_KEY
export MCP_TOOL_TIMEOUT=600000
claude mcp add --transport http pilot https://pilot.e3d.ai/mcp --header "Authorization: Bearer $PILOT_KEY"

Codex CLI

Add this to ~/.codex/config.toml and export PILOT_KEY before starting Codex.

[mcp_servers.pilot]
url = "https://pilot.e3d.ai/mcp"
bearer_token_env_var = "PILOT_KEY"
tool_timeout_sec = 600

claude.ai

In Settings, Connectors, add a custom connector. Choose No sign-in and send the key alone, with no Bearer prefix, in a request header named x-auth-token.

URL             https://pilot.e3d.ai/mcp
Authentication  No sign-in
Request header  x-auth-token: YOUR_API_KEY

Put your next plan to a debate.

Create an account See pricing