> ## Documentation Index
> Fetch the complete documentation index at: https://cyberpaisa-dof-mesh-40-27.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Autonomous Operations

> MeshDaemon, autonomous planner, auto-scaling, and task routing by type.

## Autonomous Daemon

`core/autonomous_daemon.py` runs a continuous Perceive → Decide → Execute → Evaluate loop:

```bash theme={null}
python3 core/autonomous_daemon.py --multi --model claude-sonnet-4-6
# → 3 daemons in parallel
```

Each cycle:

1. **Perceive** — read environment state, pending tasks, provider health
2. **Decide** — route task to optimal node via MeshRouter
3. **Execute** — run crew or single agent with governance pipeline
4. **Evaluate** — Supervisor scores result; retry or escalate if needed

Cycle history: `logs/daemon/cycles.jsonl`

***

## Task Routing

The mesh routes by task type to the primary node, with fallbacks in order:

| Task type  | Primary         | Fallbacks                              |
| ---------- | --------------- | -------------------------------------- |
| `code`     | deepseek-coder  | minimax → sambanova-llama → nvidia-nim |
| `tests`    | sambanova-llama | cerebras-llama → groq-llama            |
| `docs`     | cerebras-llama  | gemini-flash → groq-llama              |
| `analysis` | minimax         | cerebras-llama → groq-llama            |
| `security` | dof-guardian    | local-agi-m4max → nvidia-nim           |

***

## Supervisor Scoring

Every task output is scored before acceptance:

```
score = Q(0.40) + A(0.25) + C(0.20) + F(0.15)
```

| Score   | Verdict    | Action                         |
| ------- | ---------- | ------------------------------ |
| ≥ 7.0   | `ACCEPT`   | Output accepted                |
| 5.0–6.9 | `RETRY`    | Re-run with different provider |
| \< 5.0  | `ESCALATE` | Halt task, notify Soberano     |

Max retries: 2 (configurable in `dof.constitution.yml`)

***

## Auto-Scaling

`core/mesh_orchestrator.py` computes a scaling decision every cycle:

```python theme={null}
D_net = 0.6 × (queue_depth / 50) + 0.4 × (avg_latency / sla_ms)

if D_net > 1.0:  scale_up
if D_net < 0.2:  scale_down
else:            hold
```

***

## MeshDaemon Configuration

```python theme={null}
from core.autonomous_daemon import AutonomousDaemon

daemon = AutonomousDaemon(
    model="claude-sonnet-4-6",
    max_cycles=100,
    cycle_interval_seconds=30
)
daemon.run()
```

Session state is persisted via `core/session_resume.py` — cycle count and
improvements survive restarts.

***

## Claude Commander — 5 Modes

```bash theme={null}
python3 core/claude_commander.py
```

| Mode   | Description                    |
| ------ | ------------------------------ |
| SDK    | Direct Claude SDK calls        |
| Spawn  | Spawn worker agents            |
| Team   | Agent team with shared context |
| Debate | Multi-agent debate             |
| Peers  | Peer-to-peer coordination      |

***

<CardGroup cols={2}>
  <Card title="Mesh Nodes" icon="network-wired" href="/concepts/mesh-nodes">
    Node table and circuit breaker
  </Card>

  <Card title="Running the Mesh" icon="play" href="/guides/running-the-mesh">
    Launch commands
  </Card>
</CardGroup>