Open-source SDLC harness cuts Claude Code bug-fixing costs by up to 75% on large repos
NeighborhoodOwn8510 · reddit · 2026-07-26
What it does
AutoDev Studio is an open-source SDLC harness that takes a Git repo and a plain-English change request, then runs a multi-agent pipeline: clarify requirements, human approval, implementation on an isolated branch, real test execution, cross-family review, bounded repair loops, and finally a pull request for human merge.
Reported results
The author benchmarked it on two large Python repos, around 35k and 82k LOC. On six well-localized tasks, the tuned pipeline beat a cold single-agent Claude Code run every time, coming in 7% to 75% cheaper. One example: a cold Claude Code run spent $6.83 and 207 turns hunting a bug in an 82k-line repo, while AutoDev Studio fixed the same bug for about $1.70.
Where it loses
The author is explicit that the approach is not always cheaper:
- On trivially greppable one-line edits, the five-stage overhead can cost more than it saves.
- On one hard cross-cutting bug, it produced a cheaper but narrower fix.
How it avoids staleness
The retrieval layer is not just a vector index. It uses a deterministic symbol map synced to the latest commit at run start, plus live grep against the current working tree and the current file contents in the prompt. Embeddings only affect ranking and are refreshed incrementally per changed module.
Other details
- Model/provider agnostic per stage, including Anthropic, Claude Code CLI, OpenAI-compatible endpoints, Groq, Gemini, xAI, OpenRouter, and local Ollama
- Can run on free tiers with Groq plus a local embedding model
- Language-agnostic, with Python parsed exactly and other languages handled by lighter extractors
- Tracks real per-ticket cost, tokens, and duration
- MIT licensed, local-first, with encrypted API keys and demo mode for dry-running PRs
Related event: Open-Source AutoDev Studio Claims Up to 75% Cost Reduction(2 posts)→
More from coding & agent
- RTK is a Rust CLI proxy that cuts AI agent shell output by up to 90% — OpenAIDevs · 2026-07-26
- Daniel Lockyer recommends Moshi and herdrdev for remote agent coding and multiplexing — DanielLockyer · 2026-07-26
- ChatGPT Voice reportedly completed a Meta Business Suite flow hands-free — OpenAIDevs · 2026-07-26
- MCP server builders are debating how to scope agent permissions and API keys — Practical-Title7385 · 2026-07-26
- Open-source project targets the 90% of terminal output that coding agents waste context on — thetripathi58 · 2026-07-26
- RTK: Open-Source CLI Proxy Compresses Terminal Noise for AI Coding Agents — thetripathi58 · 2026-07-26