Oriel — AI & technology worth noticing, every day.

Memdoor: A team of 5 Qwen3.8-27B models beats Fable 5 on LiveCodeBench Hard

Repository · https://github.com/guregodevo/memdoor

First seen by Oriel on .

AI summary

Memdoor is an Apache 2.0, headless-first coding and automation agent that unifies local, SSH, editor, CI, cron, and phone-driven work through persistent gateway conversations.

  • It ships as one static binary for macOS, Linux, and Windows, requiring no Docker or Python and supporting provider keys, subscriptions, or local models.
  • Repository workflows use YAML task graphs with parallel steps, executable checks, approval gates, and resumable failures, rather than trusting model-reported completion.
  • Its Jev decision model filters context and tools; across 44 paired runs, all passed, while one codebase question fell from 54,661 to 14,367 input tokens.
  • The GitHub Action includes nine workflows for pull-request triage, verified reviews, security checks, test generation and repair, failure explanation, issue triage, descriptions, and release notes.
  • Files, sessions, and commands remain on the gateway machine; model-read content goes to the configured provider, while local runtimes can keep processing on-device.

#coding agents #workflow automation #cli #local llm

Generated by gpt-5.6-sol. Check the original for details.

Metric trends

Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained

Points

Score reported by the source.

  • Hacker News: 454, latest observation ; +440 over observed period since

Comments

Comment count reported by the source.

  • Hacker News: 229, latest observation ; +228 over observed period since

Aliases

SourceExternal IDURL
Hacker News50320995https://github.com/guregodevo/memdoor/tree/main/examples/workflows/team-solve Read discussion