Why LazyDev is for you

Four reasons teams keep it installed: it runs itself safely, its fixes are grounded instead of guessed, security is built in with System One guard + pipeline gates, and it learns your repo over time.

01

Let it run itself

From issue to PR while you review

Install the GitHub App on the repos you want monitored. When an issue is opened or reopened, LazyDev picks it up automatically — identifies the problem, researches the codebase, writes the fix, validates it in an isolated sandbox, and opens a pull request. You review and approve. It never merges on its own.

  • Automatic: opened / reopened issues are queued, no manual trigger needed
  • Queue system: one job at a time in FIFO order so fixes never overlap or corrupt each other
  • Urgent issues can jump the queue but never interrupt a running job
  • Retries with backoff on transient failures; duplicates are deduped
  • You stay in control: every change arrives as a traceable branch + PR
02

Grounded fixes, less hallucination

Serena MCP + RAG + best harness

Instead of guessing from raw file dumps, LazyDev combines semantic code retrieval (RAG over Qdrant) with precise symbol-level understanding (Serena MCP over LSP). RAG answers “where should I look?” — even when names differ. Serena answers “what exactly is there?” — signatures, callers, types — and edits exactly one symbol at a time.

  • RAG finds semantically related files + similar past fixes, not just keyword matches
  • Serena reads symbol trees and caller maps instead of whole 500-line files (~96% fewer tokens)
  • Symbol-level edits (modify / create / delete / rename) with read-before-edit guards
  • Sandbox validation in an isolated Docker container with self-healing retries
  • Full LLM logging per agent so you can see why it chose an approach
03

Security built in

System One guard + pipeline gates

Every issue passes through a free Clef model on Cloudflare Workers AI (System One) that screens for prompt injection and exfiltration attempts before the main agent sees it. Three pipeline gates — triage, patch safety, validation — each quarantine risky jobs with clear verdicts. Human approval from the dashboard or MCP chat unblocks instantly.

  • Clef on Workers AI free tier (10K Neurons/day) — no card, no surprise bills
  • Two-layer screening: regex patterns (always on) + Clef decision model (optional)
  • Triage gate classifies issue kind, complexity, and whether human review is needed
  • Patch gate detects dependency changes, protected paths, and security-sensitive edits
  • Validation gate classifies failure types (test, type, lint, env, missing dep) and routes accordingly
  • Gate verdicts on run detail: stoppedBy, decision, risk level, plain-words reason
04

Gets smarter the longer you use it

Serena memories compound

LazyDev keeps two long-lived project memories: the central repository structure and conventions, and the issues faced plus approaches tried — what worked and what did not. Each run reads them, each successful run updates them, and they are versioned alongside your code so future issues get better decisions.

  • global_repo_structure: layout, conventions, build setup for your repo
  • historical_issues_and_lessons: past bugs, approaches tried, what worked
  • Memories persist across jobs in long-term storage, not the temporary worktree
  • Recent lessons are folded into every new fix plan automatically
  • Versioned in .lazydev/memory/ inside the fix PR so the whole team can see them

Want the technical detail?

Under the hood it is a multi-agent pipeline (Onboarding → Planner → PatchGenerator → Validation → Git) with BullMQ + Redis queue, Redlock per repo/branch, Qdrant RAG, Serena MCP sidecar, and Docker sandbox validation. Three pipeline gates (triage, patch, validation) and a System One guard (Clef on Cloudflare Workers AI) quarantine risky jobs with gate verdicts shown on run detail. You can also drive it manually from chat with MCP tools liketrigger_issue_fix, implement_new_feature, get_pipeline_status, and provide_human_feedback.

Ready to pick your deployment?

Self-host for free, or let us run it for you.

Compare both options