Skip to content

mikan

Multi-platform AI coding agent for Slack, Telegram, Discord, and GitHub.

AI coding agent · pre-1.0

One agent, every chat.

mikan is an AI coding agent that lives in Slack, Telegram, Discord, and GitHub — and keeps the chat record, its own session, the sandbox it runs in, and your credentials each in their own place.

Runs where your team already talks
  • Slack
  • Telegram
  • Discord
  • GitHub

Multi-platform

One agent across Slack, Telegram, Discord, and GitHub. Mention it in a channel or DM it directly.

Concurrent sessions

Slack threads, Discord replies, and Telegram reply chains each run as an independent session.

Sandbox execution

Run tool commands on the host, in a shared container, in a per-conversation container, or a local microVM.

Credential vaults

Store keys with /login. mikan injects them as env and mounted files only at run time.

Persistent memory

A workspace-level MEMORY.md plus one per office, which the agent reads and writes as it works.

Skills & events

Drop CLI tools into skills/, and schedule one-shot or recurring work with JSON events.

Separation of concerns

Four layers on one path — kept apart on purpose.

Every conversation is an office: its own directory, its own sandbox, its own credentials. Inside one, mikan still keeps the transcript, the model loop, and the shell separate, so each can be inspected, swapped, or locked down on its own.

01 · feeds →

Chat record

The platform-facing history — log.jsonl, attachments, and files, inside that office's directory.

02 · drives →

Agent session

Structured turns with compaction and retry, persisted under sessions/.

03 · unlocks →

Sandbox

Where tool commands actually run: host, container, image, or microVM.

04

Vault

Runtime credentials as env vars and mounted secret files — never at rest in the repo.

How it works

Install it. Point it at a chat. Go.

  1. Install

    One npm install, supervised by PM2 so it restarts on failure and starts on boot.

  2. Connect

    Set a bot token per platform in mikan.env — run as many platforms at once as you like.

  3. Talk to it

    @mention it in a channel or send a DM. Each thread becomes its own agent session.

Put an agent in your team's chat.

Free and MIT-licensed. Bring your own model provider and deploy in minutes.