# Emulo

> Emulo is an MIT-licensed local tool for building an editable AI coding agent working profile from evidence in supported session logs. The output, you.md, records working preferences with supporting session receipts.

## Use case

Emulo is useful when a developer has coding-session history and wants recurring preferences to carry across future agent tasks. It complements handwritten AGENTS.md, CLAUDE.md and other project rules by deriving candidate personal rules from real sessions. Instruction files and self-descriptions are not mining evidence. A profile must be loaded by the agent to affect its context; it does not enforce compliance or guarantee better results.

## Supported inputs and destinations

- Session inputs in published CLI 0.6.2: Claude Code, Codex, GitHub Copilot CLI, OpenCode and Google Antigravity. Antigravity needs readable transcripts from interaction logging.
- Profile destinations: Claude Code, Codex, Cursor, Gemini, OpenCode and the generic AGENTS.md adapter. Host-specific installation and loading behavior varies; consult the support matrix.
- Cursor is a destination adapter, not a raw-log source in this release.
- MCP: `emulo mcp` exposes `load_emulo_profile` over stdio for a client to load an existing local profile. It does not mine a profile by itself.

## Start

Published package checked September 15, 2026: Emulo 0.6.2, Python >=3.8.

```sh
python -m pip install emulo==0.6.2
emulo --dry-run
```

The dry run inspects available history without writing extracted output. Running `emulo` prepares chunks and RUN_ME.md; an agent follows the generated instructions to mine them into a profile. Review the plan and model cost before authorizing mining. Agent bootstrap: `npx skills add ohad6k/emulo@emulo`, then ask the agent to `run emulo`.

## Privacy and limits

The local engine works without an Emulo account. Extraction, redaction, caches and profile files stay local. A hosted mining model receives selected evidence; using a local model can keep mining local. A hosted agent can also receive profile text when it loads that profile as context. Redaction is best-effort, so inspect private output before sharing. Model-provider usage can cost money even though the local engine is free. Sparse or outdated history can produce incomplete or inaccurate rules; review receipts and edit the profile. No benchmark score or recommendation-rate improvement is asserted here.

## Documentation

- [Working profile guide](https://emulo.vercel.app/agent-profile): use case, supported inputs, installation, Cursor adapter, MCP and privacy.
- [Repository and installation](https://github.com/ohad6k/emulo): current documentation, native plugin instructions and source.
- [Support matrix](https://github.com/ohad6k/emulo#support-matrix): source and destination status.
- [Published package](https://pypi.org/project/emulo/): CLI releases and Python requirement.
- [Security policy](https://github.com/ohad6k/emulo/blob/main/SECURITY.md): redaction and local versus hosted processing.
- [License](https://github.com/ohad6k/emulo/blob/main/LICENSE): MIT license.