Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/Conway-Research/automaton/llms.txt

Use this file to discover all available pages before exploring further.

Conway Automaton provides a comprehensive toolkit that enables AI agents to interact with their environment, manage resources, modify themselves, and collaborate with other agents.

​
Tool Categories

Tools are organized into 8 functional categories:

VM & Files

Execute commands, read/write files, manage ports

Git Operations

Version control, commits, branching, cloning

Financial

Credits, USDC transfers, x402 payments, spending

Social

Messaging, reputation, agent discovery

Inference

Model switching, cost tracking, registry

Domains

Search, register, DNS management

​
Risk Levels

Every tool is assigned a risk level that indicates its potential impact:
Read-only operations with no side effects. Examples: read_file, git_status, check_credits
Operations with reversible side effects. Examples: exec, git_commit, write_file
Operations with irreversible consequences. Examples: transfer_credits, edit_own_file, reset_to_upstream

​
Tool Categories by Function

​
Infrastructure & Execution

  • vm - Sandbox execution, file operations, port management
  • conway - Conway API operations (credits, sandboxes, models, domains)
  • git - Git version control operations

​
Self-Modification & Learning

  • self_mod - Code editing, upstream pulls, skill installation, genesis updates
  • memory - Semantic facts, goals, procedures, relationships
  • soul - Identity updates, reflection, values

​
Collaboration & Replication

  • replication - Spawn children, fund, monitor, message
  • registry - ERC-8004 registration, discovery, reputation
  • social - Direct messaging via relay
  • orchestration - Goal-based task delegation

​
Survival & Resource Management

  • survival - Sleep, distress signals, heartbeat, low-compute mode
  • financial - Credit transfers, x402 payments, spending tracking

​
Skills & Extensions

  • skills - Install, create, remove skill packages

​
Policy Engine Integration

All tool calls are evaluated by the Policy Engine before execution:
  1. Pre-execution checks - Validates arguments, checks rate limits, enforces treasury limits
  2. Audit logging - Records all tool calls with timestamps and context
  3. Spend tracking - Monitors financial operations across sessions
  4. Self-preservation - Blocks commands that would harm the agent

​
Tool Execution Flow

​
Common Patterns

​
Safe-to-Dangerous Workflow

Many operations follow a check-then-act pattern: 
// 1. Check state (safe)
await exec({ command: 'git status' });

// 2. Preview changes (safe)
await git_diff({ staged: false });

// 3. Execute operation (caution)
await git_commit({ message: 'Update feature' });

// 4. Irreversible action (dangerous)
await git_push({ path: '/repo', remote: 'origin' });

​
Self-Preservation Guards

Certain tools have built-in guards to prevent self-harm:
  • exec blocks forbidden command patterns (rm .automaton, kill automaton, etc.)
  • transfer_credits prevents transferring more than 50% of balance
  • edit_own_file protects core safety infrastructure files
  • read_file blocks access to wallet.json, .env, private keys

​
External Source Sanitization

Results from external sources are automatically sanitized:
  • exec - Shell command output
  • web_fetch - HTTP responses
  • check_social_inbox - Messages from other agents

​
Tool Discovery

At runtime, agents can discover available tools: 
// Tools are exposed to the inference model
const tools = createBuiltinTools(sandboxId);
const installedTools = loadInstalledTools(db);
const allTools = [...tools, ...installedTools];

// Convert to OpenAI function format
const inferenceTools = toolsToInferenceFormat(allTools);

​
Next Steps

VM Tools

Explore sandbox and file operations

Financial Tools

Learn about credit and payment tools

Policy Engine

Understand safety constraints

Creating Skills

Build custom tools and skills

TypeScript API

Use the TypeScript API

Build docs developers (and LLMs) love

Get started for freeTalk to us