> ## Documentation Index
> Fetch the complete documentation index at: https://comis-fix-skill-import-vetting-gate.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# FAQ

> Frequently asked questions about Comis

Answers to the most common questions about running and using Comis.

## General

<AccordionGroup>
  <Accordion title="What is Comis?">
    Comis is an open-source, self-hosted, security-first runtime for AI agents that learn and act across sessions.
    It starts work from schedules, messaging channels, APIs, or the web, keeps
    important run state outside the model's prompt, applies configured access
    and cost rules, and records bounded evidence for later review. Controls
    apply only to paths that run through them.
  </Accordion>

  <Accordion title="Is Comis free?">
    Yes, Comis is fully open source under the Apache-2.0 license and free to use. There is no license fee, no usage cap, and no paid tier.

    Model providers and connected services may charge separately. A local model
    can run without a hosted-provider API credential.
  </Accordion>

  <Accordion title="What AI providers does Comis support?">
    Comis supports:

    * **Anthropic** (Claude models)
    * **OpenAI** (GPT-4 and newer)
    * **Any OpenAI-compatible provider** — this includes Ollama (for local models), Together AI, Groq, and many others

    You can configure multiple providers and assign different models to different agents. See the [Models](/agents/models) page for details.
  </Accordion>

  <Accordion title="Do I need an API key?">
    You need a usable model, but not necessarily an API key. Choose a hosted
    provider credential, a supported OAuth login, or a running local provider
    such as Ollama. See [Configuration](/installation/configuration).
  </Accordion>
</AccordionGroup>

## Cost and Resources

<AccordionGroup>
  <Accordion title="How much does it cost to run?">
    Comis itself is free. Your costs come from AI provider API usage:

    Provider cost depends on model, tokens, tool loops, media generation, and
    scheduled work. Comis cannot promise a per-message or per-execution price;
    inspect your provider billing and use Comis's cost reports and budgets.

    Comis includes token limits, graph budgets, circuit breakers, and opt-in
    spend ceilings. A configured spend ceiling can warn or refuse later model
    calls after recorded priced usage crosses the limit; it does not cancel a
    provider call that is already in progress. See [Agent Safety](/agents/safety)
    for details.
  </Accordion>

  <Accordion title="What hardware do I need?">
    Use Linux for production or macOS for development. npm installations
    require Node.js 22.19.0 or newer, Python 3, and native build tools. Capacity
    depends heavily on browser/media features, local models, traffic, and data
    retention. See [Requirements](/installation/requirements).
  </Accordion>

  <Accordion title="Can I run it on my laptop?">
    Yes. macOS and Linux are supported for development and personal use. Install Node.js 22.19.0 or newer, clone the repository, build, and start.

    For always-on operation (agents responding 24/7), use a server or VPS so the agents keep running when your laptop is closed. See the [Installation](/installation) guide to get started.
  </Accordion>
</AccordionGroup>

## Setup and Platforms

<AccordionGroup>
  <Accordion title="Which chat platforms are supported?">
    Comis supports 10 chat platforms:

    * **Telegram** — full support including groups, inline keyboards, and media
    * **Discord** — servers, DMs, threads, reactions, and embeds
    * **Slack** — workspaces with Socket Mode or HTTP mode
    * **WhatsApp:** via the Baileys WhatsApp Web client
    * **Signal** — via signal-cli (self-hosted)
    * **iMessage** — macOS only, via AppleScript integration
    * **LINE** — via the LINE Messaging API
    * **IRC** — any IRC network
    * **Email** — via IMAP/SMTP (any email provider)
    * **Microsoft Teams** — 1:1 chats, group chats, team channels/threads, Adaptive Cards

    See the [Channels](/channels) overview for a detailed comparison of features and limitations across platforms.
  </Accordion>

  <Accordion title="Can I connect multiple platforms at once?">
    Yes. Comis connects to all your configured platforms simultaneously. You can have the same agent responding on Telegram, Discord, and Slack at the same time, or assign different agents to different platforms.

    Agent [routing](/agents/routing) controls which agent responds on which channel. You can route by platform, channel, or even specific users.
  </Accordion>

  <Accordion title="How do I update Comis?">
    Pull the latest code and rebuild:

    ```bash theme={}
    git pull
    pnpm install
    pnpm build
    ```

    Then restart the daemon:

    ```bash theme={}
    pm2 restart comis
    ```

    Or with systemd:

    ```bash theme={}
    sudo systemctl restart comis
    ```

    Your configuration and data are preserved across updates — only the application code changes.
  </Accordion>
</AccordionGroup>

## Security and Privacy

<AccordionGroup>
  <Accordion title="Is my data private?">
    Comis stores its primary state locally, but data can leave the host through
    the model provider, chat channels, web/media tools, MCP servers, proxies,
    and optional observability integrations you configure. A local model alone
    does not make the whole deployment offline. Review each enabled integration
    and its data policy.
  </Accordion>

  <Accordion title="Can agents access my files?">
    The default tool-policy profile is `full`, so do not assume least privilege.
    Configure a narrower [tool policy](/skills/tool-policy) and review all
    allowed paths before exposing an agent to untrusted input.

    Node.js permissions are optional and off by default. `system.exec` attempts
    a separate OS sandbox, but can fall back to ordinary execution when its
    provider is unavailable. MCP and in-process tools are outside that exec
    sandbox. See [Security](/security).

    When `security.permission.enableNodePermissions` is enabled, the Node.js permission model also disables fd-based fs APIs (`fsync`, `fchmod`, `fchown`) in the daemon process itself — credential file writes become best-effort durability and trace permissions are best-effort. This is Linux-only and off by default. See [Node Permissions — Production fd-API Disablement](/reference/node-permissions#production-fd-api-disablement-linux) for details.
  </Accordion>

  <Accordion title="What if an agent goes rogue?">
    Comis has multiple safety layers to prevent runaway agents:

    * **Budget limits** cap token usage and estimated cost per agent
    * **Circuit breakers** automatically pause agents that hit error thresholds
    * **Tool policies** restrict which tools each agent can use
    * **Approval workflows** can require human approval when explicitly enabled

    Approvals are disabled by default and the default tool policy is broad.
    Configure limits and permissions for your threat model; no control makes a
    highly privileged agent risk-free. See [Agent Safety](/agents/safety).
  </Accordion>
</AccordionGroup>

## Related Pages

<CardGroup cols={2}>
  <Card title="Troubleshooting" icon="wrench" href="/operations/troubleshooting">
    Solutions to common issues with exact error messages
  </Card>

  <Card title="Installation" icon="download" href="/installation">
    Getting Comis installed and running
  </Card>

  <Card title="Channels" icon="comments" href="/channels">
    Connecting chat platforms
  </Card>

  <Card title="Agent Safety" icon="shield" href="/agents/safety">
    Budget limits, circuit breakers, and cost controls
  </Card>
</CardGroup>