Skip to main content

Crate selfware

Crate selfware 

Source
Expand description

§Selfware — Your Personal AI Workshop

An agentic coding harness for local LLMs. 70+ tools, multi-agent swarm, evolution engine, hooks, MCP, LSP, and a fox mascot — all local-first, no cloud required.

§Core Features

  • 70+ Tools: File, git, cargo, search, shell, PTY, browser automation, computer control, LSP, MCP, knowledge graph, vision, FIM editing
  • Multi-Agent Swarm: Up to 16 concurrent agents with consensus voting
  • Safety: Multi-layer validation, path protection, command filtering, JSONL audit logging, permission grants with expiry
  • Hooks: Event-driven automation (PreToolUse, PostToolUse, Stop)
  • MCP: Both client (connect to external tools) and server (expose selfware tools to other AI systems)
  • LSP: Semantic code intelligence via rust-analyzer, pyright, tsserver, gopls
  • Doctor: System dependency checker (30+ tools) and LLM backend diagnostics
  • Templates: Project scaffolding for Rust, Python, Node.js with QA profiles
  • Cognition: PDVR cycle (Plan-Do-Verify-Reflect), working memory
  • Persistence: Checkpoint system for long-running tasks with session resume
  • Self-Healing: Error classification, recovery strategies, exponential backoff
  • Evolution Engine: Recursive self-improvement with SAB-based fitness
  • Local-First: Runs entirely on your hardware, zero data egress

§Quick Start

ⓘ
use selfware::{Agent, Config};

let config = Config::load(None)?;
let mut agent = Agent::new(config).await?;
agent.run_task("Add unit tests for the auth module").await?;

§CLI

selfware chat                    # Interactive mode
selfware run "add unit tests"    # One-shot task
selfware doctor                  # Check system dependencies
selfware init                    # Project scaffold wizard
selfware multi-chat -n 4         # Multi-agent swarm
selfware mcp-server              # Run as MCP server

See selfware.design for full documentation.

Re-exports§

pub use safety::redact;
pub use safety::sandbox;
pub use safety::threat_modeling;
pub use analysis::analyzer;
pub use analysis::bm25;
pub use analysis::vector_store;

Modules§

agent
analysis
Code analysis module
api
checkpoint
Task Checkpointing & Persistence
cli
Selfware Workshop - Your Personal AI Companion
cognitive
Cognitive and AI capabilities module
computer
Computer control module for desktop automation.
config
Configuration Management
container
Container Management System
doctor
Selfware Doctor — system diagnostics and dependency checker.
errors
hooks
Hook system for extensible event-driven automation.
input
Modern Input System for Selfware
interview
Interview / Clarification Mode
llm_doctor
LLM Doctor — diagnostic module for local LLM backend configuration.
lsp
Language Server Protocol (LSP) client for semantic code intelligence.
mcp
Model Context Protocol (MCP) client implementation.
memory
Agent Memory Management
multiagent
Multi-Agent Chat System
output
Output Control Module
planning
Hierarchical Planning System
process_manager
Process Manager - Background Process Lifecycle Management
resource
Resource management for CPU, GPU, memory, and disk
safety
Security and safety module
supervision
Process supervision and recovery mechanisms
swarm
Multi-Agent Swarm System
telemetry
Telemetry & Observability
templates
Template engine for scaffolding new projects from embedded templates.
token_count
Shared token counting utilities.
tokens
Token Usage Tracking and Cost Optimization
tool_parser
Robust tool call parser with XML and JSON fallback
tools
ui
Selfware UI System
verification
Verification Gates - Automatic validation after every code change
workflows
Agent Workflows

Macros§

tool_span
Create a span for tracking tool execution with automatic duration and outcome logging

Functions§

is_shutdown_requested
Check whether a graceful shutdown has been requested.
request_shutdown
Signal that a graceful shutdown has been requested.
shutdown_tracing
Flush and shut down the tracing background writer. Call this during graceful shutdown to ensure all logs are flushed.