MAXIM
Operating Modes
Three Autonomy Levels + Sleep as a Tool
Maxim's behavior is controlled by two independent dimensions: how much it processes (awake or sleeping) and how much authority it has (passive, active, or singularity). Sleep is not a mode—the agent enters it by calling the sleep tool, and wakes automatically when user input arrives.
The Two-Axis Architecture
Maxim's state is tracked by two independent axes:
Sleep is a processing state the agent enters by calling the sleep tool. A sleeping agent retains its operational mode and wakes automatically when user input arrives, resuming exactly where it left off.
Processing State
Controls resource usage
Operational Mode
Determines action authority
Autonomy Levels
The three operational modes control how much authority Maxim has. Each level sets a maximum initiative value between 0.0 (fully reactive) and 1.0 (fully proactive).
planning Planning Autonomy
Goal: Observe, understand, and propose actions without unilateral execution.
Max Initiative: 0.3 (mostly reactive)
| Permission | Access |
|---|---|
Sandbox (.maxim_sandbox/) | Always writable |
| CWD files | Read only, edits require approval |
| Code execution | Not allowed |
| Network | Allowed |
Forbidden tools: execute_file, maxim_command, request_directory_change
active Supervised Autonomy
Goal: Execute tasks and take actions within defined boundaries.
Max Initiative: 0.7 (proactive within bounds)
| Permission | Access |
|---|---|
| Sandbox | Full read/write |
| CWD files | Read + suggest edits (shown for approval) |
| Code execution | Requires approval |
| Network | Allowed |
No forbidden tools (execution gated by approval)
autonomous Full Autonomy
Goal: Operate with full autonomy, self-correcting and learning continuously.
Max Initiative: 1.0 (fully proactive)
| Permission | Access |
|---|---|
| Sandbox | Full access including execution |
| CWD files | Full read/write/execute |
| Code execution | Allowed |
| Network | Allowed |
No forbidden tools, full tool access
Safety and ethical constraints (Constitution) still apply unconditionally
Sleep
Sleep is a processing state, not a mode. The agent enters sleep by calling the sleep tool. It retains its operational mode but dramatically reduces processing:
awake Full Processing
- Full LLM processing active
- All tools available (per mode constraints)
- Default Network (orienting, social) enabled
- Video and audio capture running
sleep Background Only
- LLM processing skipped
- Background tasks: memory consolidation, pattern extraction
- Only
respondtool available - Default Network disabled
- Wakes automatically on user input
Biological Parallel
Like biological sleep, Maxim's sleep state isn't unconsciousness. It's active maintenance: consolidating memories, extracting patterns, cleaning up. The audio monitoring is analogous to the brain's ability to detect your name even while sleeping.
Headless Mode
When no robot hardware is detected, Maxim automatically enters headless mode — the full agentic loop runs without media capture, motor control, or Default Network overhead. Detection uses mDNS: if the robot's hostname doesn't resolve within 5 seconds, Maxim skips the SDK connection and starts immediately.
headless No Robot Connected
- Full LLM processing and agentic loop active
- CLI input and file-change perception only (no video/audio)
- Default Network disabled (no motor commands)
- Frame/audio capture workers not started
- All bio memory systems active (Hippocampus, ATL, NAc, AG, EC)
- Tools, provenance, and concept memory fully operational
Set MAXIM_ROBOT_TIMEOUT=5 to reduce the mDNS timeout for faster headless startup. The system uses capability detection, not mode flags — like a biological organism adapting to sensory loss rather than requiring an explicit switch.
Mode Switching
Maxim supports multiple mechanisms for changing state at runtime:
Voice Commands
- "Maxim sleep" → enter sleep (calls sleep tool)
- "Maxim wake up" → awake processing state
- "Maxim planning" → planning operational mode
- "Maxim active" → active operational mode
- "Maxim autonomous" → autonomous mode
Agent Tools
The agent can change modes through tools:
mode_switch— Switch between operational modes. Logs switches with timestamps and reasoning.autonomy_level— Request autonomy changes. Escalation is always allowed. De-escalation requires human approval via a proposal queue.sleep— Enter the sleep processing state. The agent wakes automatically when user input arrives.