> ## 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.

# Sage

> Sage is an AI operator that lives inside your Mythic C2: it reads your live operation, reasons like an analyst over the BloodHound attack graph, and drives real Mythic tasks.

You work with Sage in a [Mythic](https://github.com/its-a-feature/Mythic) chat channel. It reads
the callbacks, tasks, credentials, and files in your operation alongside your BloodHound graph.
Ask it a scoped question about the engagement and get an answer grounded in that data, or hand it
an objective and let it plan and drive the path itself.

> [!WARNING]
>
> **Authorized use only, and still maturing.** Use Sage only for activity you are explicitly
> authorized to perform. Sage is in beta: an autonomous agent driving a C2 carries real risk of
> unintended impact on a live network. Weigh production use very carefully, and prefer labs and
> authorized engagements until Sage has matured.
>
> Auto mode is the sharp edge of that risk, and it is not the only way to run Sage. Supervised
> mode is human-in-the-loop: Sage proposes, and every guarded action waits for your explicit
> approval before it runs. Conversation mode is read-only. Start there.

## Scope and boundaries

Sage installs into a Mythic server the same way a payload type or C2 profile does, and it has no
function outside one. It runs on the control plane and operates through your Mythic payloads.
It is not an implant, and it does not create a Sage payload callback. Sage is a native Mythic
**chat container**, so it requires Mythic v4: chat containers do not exist before v4. The
beta-qualified target is **Mythic `v4.0.0rc5`**; newer v4 builds may work but are outside the
validated matrix. See [Limitations and validated configuration](https://docs.mythicsage.ai/understand/limitations.md).

Three boundaries define the product:

* **Target-facing activity always executes through a live Mythic payload callback.** Sage may
  query Mythic and BloodHound control-plane data, but it never connects directly to target LDAP,
  SMB, Kerberos, WinRM, RPC, HTTP, or similar services.
* **Sage-local attack artifacts are not proof.** Runtime effects and objective completion require
  Mythic task output and artifacts, Mythic credential-store state, or BloodHound facts derived
  from payload-collected artifacts. Self-reported success is not recorded.
* **Proof is enforced, not asserted.** A durable [engagement
  ledger](https://docs.mythicsage.ai/understand/how-sage-works.md#the-engagement-ledger) records each achieved effect with a
  proof envelope: the Mythic callback, task, and transaction ids, plus verifier result hashes. A
  row without a valid envelope can never count as achieved.

## Two ways to work

> **Assisted: conversation and supervised**
>
> Ask scoped questions about the operation in `conversation` mode (read-only, the safe
> default). Switch to `supervised` to let Sage propose guarded actions; every one of them waits
> for your explicit approval, default-deny, before it runs.

> **Auto: autonomous objectives**
>
> Give Sage a scoped objective and it runs the observe → select → execute → verify loop without
> per-action approval. This is experimental: its current validation is the
> [GOAD](https://github.com/Orange-Cyberdefense/GOAD) lab environment only, and it does not
> establish reliability or generalization elsewhere.

## Next steps

> **[Quickstart](https://docs.mythicsage.ai/quickstart.md)**
>
> Install Sage into Mythic and run your first conversation.

> **[How Sage works](https://docs.mythicsage.ai/understand/how-sage-works.md)**
>
> The observe → model → select → execute → verify loop, the specialist agent graph, and the
> engagement ledger.

> **[Authority modes](https://docs.mythicsage.ai/understand/authority-modes.md)**
>
> What `conversation`, `supervised`, and `auto` mean, and exactly where the approval wall sits.

> **[Security and data handling](https://docs.mythicsage.ai/understand/security-and-data-handling.md)**
>
> The control-plane boundary, evidence rules, and what Sage does with your data.


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