-
-
Notifications
You must be signed in to change notification settings - Fork 1
Home
Welcome to the comprehensive documentation for M31 Autonomous (M31A) -- the terminal-native AI coding agent that ships, not just suggests.
M31 Autonomous is a terminal-based AI coding agent written in Go. Unlike browser-bound assistants, it runs inside your shell, owns a seven-phase workflow end-to-end -- Initialize, Discuss, Plan, Execute, Verify, Runtime, Ship -- and commits verified changes to your git tree. One static binary, zero telemetry, any POSIX shell.
| Page | Description |
|---|---|
| Getting Started | Installation, first run, and basic setup |
| Configuration | All configuration options with examples |
| Slash Commands | Complete command reference |
| Keybindings | Keyboard shortcuts, leader keys, and mouse support |
| Page | Description |
|---|---|
| Architecture | High-level system design and module layout |
| Seven-Phase Workflow | The Initialize-Discuss-Plan-Execute-Verify-Runtime-Ship engine |
| Tools System | Built-in tools (19), dispatcher, permissions, and security |
| Provider System | LLM providers, model cache, fallback, retry, and SSE streaming |
| NVIDIA NIM Provider | NVIDIA NIM integration, setup, and configuration |
| Subagents | Parallel subagent system with profiles and worktree isolation |
| Task Runner | Dependency-aware task scheduler with parallel execution |
| Prompt System | Embedded prompt templates and plan parsing |
| Decision Logging | In-memory decision logger with redaction and browsing |
| Knowledge System | Cross-session project knowledge and convention detection |
| Page | Description |
|---|---|
| Code Intelligence | Source parsing, import graphs, symbol indexing, relevance scoring, EFIE backend |
| CodeComplexity Tool | Codebase complexity analysis and reporting |
| Token Estimation | tiktoken-based counting, EMA calibration, context warnings |
| Git Integration | Git operations wrapper: commits, diffs, branches, stashing |
| Page | Description |
|---|---|
| Terminal UI | Bubble Tea TUI: 33 screens, 11 themes, home screen, and layout |
| Command Palette | Full-screen searchable command browser |
| Chat History | Scrollable session message browser |
| Mouse Support | Mouse wheel, click, and scrollbar interaction |
| Session Management | Sessions, checkpoints, persistence, coordinator, and compaction |
| Prompt History | Frecent prompt history with scoring and autocomplete |
| Design Tokens | Design system: color palette, typography, spacing, elevation, motion |
| Layout Engine | Flexbox-inspired layout system with responsive breakpoints |
| Accessibility | Screen reader support and terminal detection |
| Page | Description |
|---|---|
| Public Packages | Importable pkg/ packages: ledger, rollback, bisect, arbitrage, metrics, compaction, coordinator, retry, skills |
| Metrics & Observability | Session observability pipeline: tool, LLM, and phase metrics |
| Security Model | Permission system, keychain, sandboxing, and SSRF protection |
| Error Handling | Sentinel errors and user-friendly error messages |
| Logging | Structured logging with JSON output and daily rotation |
| File Utilities | Atomic file writes for crash-safe persistence |
| Context Registry | Dynamic context injection with change detection |
| Development Guide | Building, testing, linting, and contributing |
v1.7.1 -- Architecture reorganization (internal/ui/tui, internal/engine/taskrunner, internal/integrations/context), subagent profile improvements, test infrastructure consolidation, Windows file locking and DNS cache utilities, CI fixes (Go 1.25.0, timeouts, artifact uploads), session persistence race fixes, workflow engine wiring fixes, removed dead code and deprecated packages.
v1.7.0 -- Extracted PhaseCoordinator, StateMachine, WorkflowCache, ContextBuilder, CostTracker, PromptBuilder from Engine; centralized retry logic; performance/concurrency/memory optimization helpers; audit logging; sidebar by default on wide terminals; breadcrumb navigation; keyboard standardization; visible focus indicators; text-based health status; ANSI SGR fallbacks; expanded screen reader announcements; lightweight streaming markdown parser (10fps); viewport virtualization; resize debounce; quick mode, /help getting-started, /skip commands; getting-started tour; improved home screen; improved toast system; permission modal UX; diff viewer improvements; workflow phase progress indicator; keyboard-accessible empty states; help screen and command palette sync; config externalization (hardcoded refactoring, threat model validations); provider configurable fallback priority/health checks/registration order; model capability config overrides; prompt loader with 4-level override; tool behavior externalization (dangerous commands, subagent profiles, narrative/compaction templates); UI externalization (logo, welcome, symbols, theme, website); multi-provider token estimation; empty state templates; permission descriptions; narrative engine for TUI storytelling.
v1.6.0 -- Build metadata embedding (commit, date, Go version via ldflags), -trimpath for reproducible builds, GoReleaser changelog filtering. 19 built-in tools, 33 TUI screens, 11 themes, 7 workflow phases, 3 LLM providers with automatic fallback, cross-platform support (Linux, macOS, Windows).
v1.5.0 -- Decision transparency (in-memory decision logger with redaction and browsing), cross-session project knowledge store with convention detection and growth caps, checkpoint resume with disk-backed session checkpoints, self-heal explanation (HealReport struct), file-level selective rollback, multiple race condition fixes (provider, workflow, session, ledger), TOCTOU race mitigation for port discovery.
v1.4.0 -- EFIE backend (graph analysis, bloom filters, centrality scoring, community detection, polyglot parsers), runtime verification phase with dev server lifecycle and smoke tests, home screen, subagent profiles with streaming tool calls, session coordinator with file locking, session compaction, dynamic context registry, retry policy with exponential backoff, skill discovery, persistent permissions, output store, DevServer/HTTPCheck tools, diff summaries, agent switching, post-ship validation, model prompt templates, and runtime UI. TODO cancelled status, TODO sync, NVIDIA NIM provider, CITATION.cff, security policy, enhanced input validation, FileWrite tool enforcement, improved signal handling, centralized health checks, unified UI components, updated website build instructions.