> ## Documentation Index
> Fetch the complete documentation index at: https://docs.mythicsage.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Quickstart

> Install Sage into a running Mythic server and hold your first conversation.

This page takes you from a running Mythic v4 install to your first Sage conversation.

**1. Check the requirements**

Sage is a native chat container, which only exists in Mythic v4. The beta-qualified target is
**Mythic `v4.0.0rc5`**. Earlier Mythic versions have no native chat support, and other Mythic
v4 revisions are unvalidated for this beta. Model-provider credentials or an OpenAI-compatible
endpoint are required before your first conversation.

**2. Install Sage**

From your Mythic directory, install Sage the same way as any agent:

```bash theme={null}
sudo ./mythic-cli install github https://github.com/MythicAgents/sage
```

The installer pulls the published container image named in `config.json` instead of building
one. To build from a local checkout instead, run
`sudo ./mythic-cli config set sage_use_build_context true`, then `sudo ./mythic-cli build sage`.

> [!WARNING]
>
> A reinstall replaces the whole `InstalledServices/sage/` directory, including the SQLite
> databases, the engagement ledger, and trace history in it. Archive them first if you need
> the history. See [Runtime state and resets](https://docs.mythicsage.ai/configuration/runtime-state.md).

**3. Create a Sage chat channel**

After the container registers, create a Mythic chat channel using the `Sage` model. Set the
channel `mode` here, or switch it at any time with `/mode`. Start with `conversation`: it is
read-only and the safe default.

Give the channel an operation bot's API token rather than your own, so Mythic attributes
Sage's actions to the bot. The token needs `apitoken.write` and `chat-ai.write`, plus the
write scopes for any guarded tools you want available. The scopes and every other field in
the dialog are covered in [Chat channel configuration](https://docs.mythicsage.ai/guides/chat-config.md).

**4. Configure the model provider**

Fill in provider, model, and credentials in the channel configuration, or store credentials as
Mythic user secrets. See [Environment and configuration](https://docs.mythicsage.ai/configuration/environment.md) for how
settings resolve.

**5. Start the conversation**

Ask a scoped question about your operation in chat: which callbacks exist, what the credential
store holds, what the BloodHound graph looks like. Mythic renders Sage output as markdown, with
each tool execution shown as an updating, collapsible card.

**6. Confirm it works**

Run `/tools` in the channel. It answers without a model call, so a tool list proves the
container and channel are working before provider credentials are involved; it does not yet
prove the token's scopes. Ask your first question, then check the [status chips](https://docs.mythicsage.ai/guides/using-chat.md#reading-the-chips) above the transcript:
they show the provider, model, and mode the channel actually resolved. If a step fails, see
[Troubleshooting](https://docs.mythicsage.ai/guides/troubleshooting.md).

## macOS note

If you run Mythic via Docker Desktop or OrbStack on macOS, Sage's SQLite databases sit on a
file-sharing layer that does not support SQLite locking, and Sage will report `unable to open
database file`. Move the databases inside the container with two `.env` lines. See
[Runtime state and clean resets](https://docs.mythicsage.ai/configuration/runtime-state.md#sqlite-on-macos).

## Where to go next

> **[Chat channel configuration](https://docs.mythicsage.ai/guides/chat-config.md)**
>
> Every field in the New AI Chat dialog, and where credentials resolve.

> **[Using Sage in chat](https://docs.mythicsage.ai/guides/using-chat.md)**
>
> Modes, approvals, and slash commands.

> **[BloodHound](https://docs.mythicsage.ai/guides/bloodhound.md)**
>
> Connect the graph that Sage observes and plans against.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.