Local-first · BYO-GPU · open core

Self-hosted build agent.
Cents, not credits.

Drop an idea in a text file. Get back a verified, working repo. Localsmith chains your local and cloud LLMs to ship small projects unattended — routing the free work to your own GPU and spending cash only on the one step that needs it. Your code never leaves your LAN.

# one idea in… $ echo "# Build a URL shortener in Go with a Dockerfile" > queue/short.md $ bin/orchestrate.sh queue/short.md scaffold ollama qwen2.5-coder free tree.json ✓ plan claude (Pro quota) $0.00 plan.json ✓ implement z.ai glm-4.6 $0.014 src/*, Dockerfile ✓ verify gofmt + local review free report.md ✓ docker build OK # …a working repo out. Total spend: $0.014

Four stages. One of them costs money.

A small, auditable pipeline — no LangGraph, no Temporal, no cloud control plane. Just state.json per project and a resumable bash driver.

queue/<idea>.md │ ▼ scaffold local Ollama — free → empty repo skeleton ▼ plan Claude / Codex CLI (your quota) → one task per file ▼ implement Z.AI GLM — ~1–2¢ (or fully local) → source, README, Dockerfile ▼ verify linters + local-model review — free → report.md, Docker build check ▼ shipped, verified repo

Why people self-host their build agent

💸 Cents, not credits

Cloud agents meter every token. Localsmith pushes scaffolding, planning and review to models you already run — cash is spent only on implementation. Real runs cost ~1–2¢.

🔒 Your code stays home

Prompts and source never have to leave your LAN. Run fully local, or bring your own keys. Nothing is sent to us — there is no "us" server.

🧱 Boringly auditable

Stdlib Python + bash. One JSONL cost ledger. Resumable, idempotent, 27 tests. You can read the whole thing in an afternoon.

🏠 Homelab-native

Wires straight into Ollama, LM Studio, and SearXNG. Ships with a one-command setup wizard that detects your local models.

Pricing

The engine is free and open. Pay for the parts that save you the most time.

Engine

£0

Open source · self-host

  • Full scaffold→plan→implement→verify pipeline
  • 6 providers, cost ledger
  • Setup wizard
  • Community support
Clone on GitHub

Pro founder

£99 one-time

Lifetime, for the first cohort

  • Everything in Engine
  • Self-repair loop — auto-fix failed builds
  • Provider-routing presets & prompt packs
  • Queue dashboard UI
  • Priority support
Buy lifetime (Lemon Squeezy)

Homelab Pack

£39

Add-on · one-time

  • TrueNas / docker-compose bundle
  • Ollama + SearXNG wired in
  • Stand it up on a home server in minutes
  • Runbooks from a real homelab
Add the pack