mythos-router  by thewaltero

Verifiable AI-assisted coding CLI and SDK

Created 3 months ago
274 stars

Top 94.1% on SourcePulse

GitHubView on GitHub
Project Summary

<2-3 sentences summarising what the project addresses and solves, the target audience, and the benefit.> Mythos Router provides a local CLI and SDK for verifiable AI-assisted coding, addressing AI agents' tendency to hallucinate file operations. It implements Strict Write Discipline (SWD) to ensure AI actions are trustworthy, preventing data corruption and providing tamper-evident receipts. This tool is for engineers and power users seeking reliable AI code generation with a filesystem-level trust boundary.

How It Works

SWD captures SHA-256 filesystem snapshots before and after AI-claimed file operations, verifying them against the actual state. Discrepancies trigger rollbacks, preventing partial writes. Successful operations yield hash-chained, tamper-evident receipts. This model-agnostic approach parses text-based [FILE_ACTION] blocks from any agent (Anthropic, OpenAI, etc.), establishing the filesystem as the trust boundary.

Quick Start & Requirements

  • Installation: npm install -g mythos-router or npx mythos-router chat.
  • Prerequisites: Node.js 22+ (22.5.0+ for advanced SQLite features). API keys for Anthropic, OpenAI, DeepSeek, or Surplus are required for interactive commands.
  • Setup: Configure API keys; mythos init for project onboarding.
  • Links: Quick Start: npx mythos-router chat; Onboarding: mythos init; Examples: examples/verified-writes.

Highlighted Details

  • Strict Write Discipline (SWD): Verifies AI file operations against the filesystem via SHA-256 snapshots, rolling back invalid claims.
  • Tamper-Evident Receipts: Generates hash-chained receipts for auditable trails of file operations.
  • Multi-Provider BYOK: Supports Anthropic, DeepSeek, OpenAI, and Surplus keys with automatic routing.
  • Verified Cost-Router: Escalates to expensive models only when cheaper ones are provably wrong.
  • Adaptive Thinking: Uses Claude Opus 4.8, Sonnet 4.6, and Haiku 4.5 based on --effort levels.
  • Self-Healing Memory: Reconstructs agent memory from MEMORY.md via a SQLite FTS5 index.

Maintenance & Community

The README provides no specific details on notable contributors, sponsorships, or community channels.

Licensing & Compatibility

  • License: MIT License.
  • Compatibility: Permissive MIT license allows commercial use and integration into closed-source projects.

Limitations & Caveats

Relies on parsing text-based [FILE_ACTION] blocks, noted as more fragile than native tool calls. Interactive commands require API keys. --check and --run-checks operate within user permissions, not OS sandboxing. Node.js 22+ is a strict requirement.

Health Check
Last Commit

1 day ago

Responsiveness

Inactive

Pull Requests (30d)
10
Issues (30d)
0
Star History
67 stars in the last 30 days

Explore Similar Projects

Starred by Peter Norvig Peter Norvig(Author of "Artificial Intelligence: A Modern Approach"; Research Director at Google).

vet by imbue-ai

0.7%
652
AI-powered verification for code and agent actions
Created 5 months ago
Updated 1 month ago
Feedback? Help us improve.