Ennodia. AI makes mistakes

But what one misses, another catches!

Paste this into your agent

try-ennodia.cherninlab.com
Read docs GitHub
Agents

Compatible local agents

Ennodia works through supported local harnesses. It does not provide a hosted model account or replace the subscriptions those tools use.

  • Codex
  • Claude Code
  • OpenCode
  • Kilo Code
  • Kiro
  • Cline
  • Hermes Agent
  • Antigravity
FAQ

Common questions

What is Ennodia?

Ennodia is a local MCP server for multi-agent review: your primary agent can ask other installed agent CLIs for second opinions, then compare the answers.

Why would one AI agent ask other agents for help?

Different agents miss different things. Ennodia is useful when extra model work is worth spending because quality, risk, or planning uncertainty matters.

How long does a run take?

Usually minutes, not seconds. Ennodia starts real local agent CLIs, and Compare adds a judge pass plus a synthesizer pass after child agents finish.

Do I need to know MCP?

No. You can paste the install line into Codex, Claude Code, Antigravity, or another agent and ask it to install Ennodia for you.

Does Ennodia replace my main AI agent?

No. Your primary agent stays in charge. Ennodia gives it a way to route a question to other local agents and bring their answers back into the same workflow.

Which agents can Ennodia ask?

Ennodia works through local harness adapters for agent CLIs such as Codex CLI, Claude Code, OpenCode, Kilo Code, Kiro CLI, Cline CLI, Hermes Agent, and Antigravity.

Does Ennodia expose an OpenAI-compatible API?

The separate Ennodia IO package exposes experimental local provider options and a non-streaming chat-completions subset.

Does Ennodia send my code to Ennodia servers?

No. Ennodia runs locally and does not use an Ennodia-hosted service. The child agent CLIs you choose may still call their own model providers according to their own settings.

Why not manually copy prompts between tools?

Manual copy-paste loses traceability and makes you the judge. Ennodia keeps selected agents, task IDs, failures, timing, outputs, model-led Compare, and the final answer in one run.

Can Ennodia limit spend?

Ennodia can estimate preflight input tokens and enforce local limits such as maxChildTasks and maxEstimatedInputTokens on that estimate. It cannot know provider billing or harness-internal context in advance.

Is there a benchmark?

Yes. The deterministic bug-recall benchmark is published in the docs with a dated snapshot and can be rerun from a checkout with bun run bench:bug-recall.

Does Ennodia support files and screenshots?

File and screenshot workflows depend on what each local harness can access. Include file paths and short descriptions today; better attachment staging is an important next step.

Is Ennodia only for code review?

No. Code review is a good fit, but Ennodia also works for docs audits, release checks, migrations, research checks, and design critique.

Getting Started

Install Ennodia

Send the install link to your primary agent, or add Ennodia to an MCP client directly.

Agent prompt

try-ennodia.cherninlab.com

Command

npx -y ennodia

MCP config

{
  "mcpServers": {
    "ennodia": {
      "command": "npx",
      "args": ["-y", "ennodia"]
    }
  }
}
Use cases

What Ennodia helps with

  • Stubborn bug

    Run several agents when one answer is not enough, then compare the likely root causes before editing.

  • Planning tradeoff

    Ask different models for migration plans so their biases show up before the project commits to one path.

  • Skill comparison

    Run with and without a skill such as source-grounded-audit, then compare what the skill changed.

  • Large plan review

    Split a broad proposal into focused audit slices, send them to different local agents, then synthesize the useful results.

  • Benchmark receipts

    Check deterministic bug-recall fixtures before trusting the pitch, then rerun them from the checkout.

Example prompt

Try a model-led review

This asks Ennodia to run other agents, compare disagreements, and return one next action. 

Use Ennodia to ask the best available local agents:
"Review this change before I ship it. Find the highest-risk issue, docs drift, and install confusion. Compare disagreements with a model and give me the next action."
Process

How Ennodia runs

  1. 1

    Ask

    Your primary agent sends one request.

    Agent Ennodia
  2. 2

    Route

    Available local harnesses are selected.

    Ennodia Codex Claude Antigrav
  3. 3

    Review

    Each agent answers independently.

    A B C Compare
  4. 4

    Return

    A model compares; you get the answer and trace.

    Judge Answer Trace
Interfaces

Core & MCP

  • Ennodia MCP

    Available

    The shipped interface agents use today: MCP tools for harness discovery, runs, budget checks, tasks, and Compare.

  • Ennodia Core

    Available

    The shared orchestration behavior behind MCP and the IO package: routing, task lifecycle, failure visibility, and model-led synthesis.

Budget

Spend extra model work on purpose

  • Uses local agent CLIs, installs, subscriptions, and model choices you already have as a review panel
  • Discovers supported harnesses automatically and keeps adapters thin
  • Provides a preflight input-token estimate plus hard local caps for max child tasks and estimated input
  • Tracks child task IDs, ETA, output, failures, Compare state, and durable terminal run history
  • Lets a model judge and synthesize parallel answers so the user is not the comparison engine
  • Keeps model and skill preference memory on the roadmap instead of pretending it exists
Compare

The model does the comparison

  1. Judge

    Names agreements, disagreements, missing context, unique value, and risk.

  2. Synthesizer

    Turns the original answers and judge notes into one answer the caller can use.

  3. Trace

    Keeps task IDs, failures, timing, partial output, and final status visible.

Receipts

Bug-recall benchmark

The current deterministic snapshot covers 4 TypeScript review fixtures. Ennodia parallel Compare scored 100% recall and 100% precision in fixture mode, with limitations documented beside the table.

Read the benchmark snapshot

Credits

Inspired by karpathy/llm-council

LLM Council demonstrates the pattern: ask several models independently, compare, synthesize. Ennodia brings that pattern into local agent workflows, MCP, task tracing, failure visibility, and reusable harness adapters.

Get started

Ask for second opinions

Paste the install line into your primary agent, then ask it to bring in reviewers when a task deserves more than one model.

Install GitHub