Purpose
Emulo's goal is to make your coding agent work according to your habits from the very first message of a conversation. It does not read your CLAUDE.md or rules files; instead, it directly mines the messages you have actually typed to coding agents, organizing these implicit rules that were never written down—such as what counts as "done," what you will immediately reject, how you debug, how you design interfaces, and how you write—into a local private profile you.md, and installs it in a location the agent can read.
Main Capabilities
Profile Mining: Selects evidence on demand from the logs of Claude Code, Codex, Copilot CLI, OpenCode, and Google Antigravity, generating layered files for work, design, writing, and video, so the host can load only the corresponding layer per task. Every rule comes with a dated quote from your own messages.
Usage Report: Runs before any mining, does not call a model, and only performs text matching on the messages you have typed, tallying requests you repeated verbatim three or more times in a row, messages containing wording such as "as I said" or "I told you," near-duplicate consecutive requests, and the frequency of messages beginning with a corrective tone (such as "no," or "that's wrong"). Content inside quotation marks, code blocks, lines beginning with >, or double-quoted content is not counted. Each finding prints the dated messages behind it; checks that do not meet the threshold also show their counts.
Profile Verification: emulo verify extracts every quote in the profile and searches the mined sessions for it; quotes that cannot be found are reported and the command exits with a non-zero code, indicating they may be fabricated, paraphrased, or from unmined history; quotes that rely on only a single session are flagged separately.
Share Card: The mining process generates card.json, and emulo --card <card_path> can render it as a shareable card, including the archetype, laws sorted by the number of different sessions supporting them, coverage statistics, and one sharp truth. The official recommendation is to share only the card or a brief trait, not the full profile.
Installation Targets: For Codex and OpenCode, emulo --install you.md --target agents --repo . writes the profile into the project's AGENTS.md; the official note warns that this file is usually committed and shared, so it should be excluded from version control or installed into a personal folder that is not committed. For Claude Code, emulo --install you.md --target claude installs a skill named you, but Claude Code will not reliably open it automatically; you need to enter /you or /you <task> to load it.
CLI and MCP: emulo --dry-run writes nothing; it only prints the number of sessions discovered, the number of messages, the approximate token count, the number of redacted messages, and the paths that will be written; emulo then writes the corpus, chunks, and RUN_ME.md into emulo-out/. It can be installed via uv tool install emulo, or run without installation via uvx emulo. emulo mcp runs an MCP server that uses only the standard library and itself makes no network calls, exposing the load_emulo_profile tool, which returns the activated work, design, writing, or video profile.
Native plugins: The Codex plugin provides five skills: emulo:mine, emulo:work, emulo:design, emulo:write, emulo:video; the Claude Code plugin registers the same five skills and can be installed via /plugin, or by running claude plugin marketplace add ohad6k/emulo in the terminal followed by claude plugin install emulo@emulo.
How to use
After installing the cross-agent bootstrap in Claude Code, have the agent run the mining process, pointing it at your session history. Official description: the bootstrap creates a read-only full-history mining plan; the agent must first show the cost and wait for approval before doing model work. If you do not want to go through an agent, you can install the CLI yourself, preview with emulo --dry-run, then run emulo to generate the corpus and RUN_ME.md, and then hand RUN_ME.md to your agent: the agent processes each chunk once, merges them and writes you.md, and prints the installation command.
When it applies
It is for developers who want coding agents to follow their existing personal work habits and reduce repeated explanations. Profiles are layered by work, design, writing, and video, so that different tasks load only the relevant layer.
Privacy and pricing
Emulo is MIT-licensed, free, and requires no account or login; the tool itself has nocontent that needs to be purchased. Session extraction, redaction, caching, the profile itself, and the agent adapters all stay on your machine; the only exception is mining: if you point it at a hosted model, the selected evidence is sent to that provider, while pointing it at a local model keeps everything on your machine. The official website also offers an optional paid service: manual editing of the profile you have already mined; using Emulo does not require it at all.
What has been verified and limitations
The official documentation clearly distinguishes three things: usage reports and emulo verify are only text matching against your own messages; they can only count and trace quotations, and cannot explain why; loading means the host puts the profile in front of the model, and which hosts have been observed doing this is shown in the support matrix; whether a loaded profile makes the agent work better has not yet been proven, and is a goal rather than a result. Published tests show: attaching a mined profile changed model outputs; in a 60-run study with all predictions written down in advance in August 2026, the mined profile was not distinguishable from a same-length fictional profile in any of the 9 comparisons; in a rerun in September 2026 using a 2026-07-30 profile, the mined profile was distinguished in 5 of 9 comparisons on Fable 5.1 and in 1 on Opus 5. Both tests measure whether the output changed (rounded corners, color saturation, card containers, length), not whether it got better. Regarding loading for Claude Code, Cursor, Gemini, OpenClaw, and Hermes Agent, the support matrix, the status of the Emulo Proof v1 methodology, and the details of the Reddit release anecdote, the materials already give specifics: that anecdote is a single-run example with no control and no placebo.
