Instar CLI
Instar runs a persistent agent on top of Claude Code or Codex with memory, scheduled jobs and replies over Telegram.
- 86 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- A persistent agent runs jobs without you and can read mail and messages in connected messengers.
- Telegram, WhatsApp and Slack tokens give access to your conversations.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/instarDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- A persistent agent runs jobs without you and can read mail and messages in connected messengers.
- Telegram, WhatsApp and Slack tokens give access to your conversations.
This is third-party code. Review the repository files before installing.
What it does
Instar starts a local server that holds Claude Code or Codex sessions and gives them long-term memory. Memory sits in SQLite with full-text search per topic, with rolling summaries and context re-injection after the window is compacted. The agent runs scheduled jobs, tracks its commitments and replies in Telegram, WhatsApp, iMessage and Slack with a single identity. A layer of safety checks applies: a coherence gate with a set of reviewers and an operation gate. The setup wizard configures the environment, messengers and identity files.
Who it is for. People who want an agent running on their own machine all the time, taking scheduled jobs and reachable from a phone.
Good fit when
- The agent must remember last-week agreements and stop repeating questions in every session.
- Mail or service health must be checked every hour with the result sent to Telegram.
- One agent is messaged from several channels while identity and context stay shared.
- Actions the agent promised must be tracked with reminders for overdue ones.
Not a fit when
- You need a one-off code edit: a persistent server and messengers are overkill.
- The agent cannot be given access to mail and messengers on this machine.
- You must work without the Claude Code CLI: it is required.
Example request
Set up a persistent agent with Instar, add an hourly mail check job and send the result to Telegram.Limitations
Node.js 20 or newer and the Claude Code CLI are required, along with an API key or a Claude subscription. The agent runs on your subscription, so persistent jobs spend its limits. Connecting Telegram, WhatsApp, iMessage and Slack sends conversations through those services. Some sentinels are meant for the developer environment of the project itself. Setup is the `npx instar` wizard plus `instar server start`, and the project changes quickly, so check versions and behavior against the guide on the website.
How to disable. Stop the Instar server, remove the package and revoke the messenger tokens you issued. Delete identity files and the memory database by hand.
FAQ
How do I start Instar?
Run `npx instar` for the setup wizard, then `instar server start`. After that you can message the agent from a connected messenger.
What does an Instar agent run on?
On Claude Code or Codex with your subscription or API key. The project brings no separate engine.
How does the agent remember past conversations?
Instar stores per-topic memory in SQLite with full-text search, builds summaries and feeds the needed context back after the window is compacted.
Which messengers are supported?
Telegram, WhatsApp, iMessage and Slack. The setup wizard configures channels, and the agent keeps one identity across all of them.
Is there protection against unwanted actions?
Yes, layered checks run: a coherence gate and an operation gate. Grant the agent only the access its jobs need.
Discussion
No questions or reviews yet.
Related
CLI toolsA Claude Code plugin that gives the agent a name, a personality and memory across sessions, plus chat channels.
Mycelium puts people and their coding agents into shared rooms with chat, a task board and memory that stays in the room.
One local memory for Codex, Claude Code, Cursor, Hermes Agent and OpenClaw: agents hand off tasks, review each other and refine their skills overnight
pikiloom runs Claude Code, Codex and Gemini from one web dashboard and chat apps so you can steer agents from a phone.
A local daemon assembles Claude Code, Codex and OpenCode into a team: tasks, separate worktrees, checks and merging.
Your own server gives every agent shared memory, AST-level code search and pre-commit checks through one MCP endpoint.
README in short
The project is distilled from the agent Dawn, which has been running continuously since early 2026. Installation and quick start documentation lives on instar.sh.