Local
Built from scratch for local LLM coding. The right local model codes smoothly on hardware you already own. No GPU? That's what HamrPass is for.
simplicity over complexity
A minimal, local-first coding agent.
Made for folks who'd rather code than configure. Bring your own key, run it locally, or reserve a HamrPass seat. No subscription, no login.
~/projects/api · codehamr
What is codehamr?
- A reliable coding agent for your terminal.
- A single config.yaml for your whole setup.
- An agent that proves it instead of promising it.
What it isn't.
- No plugin zoo to babysit.
- No login, no subscription.
- No telemetry.
pick the right tool
When to pick what.
frontierClaude Code, Cursor or Codex are great first-party agents, but they're expensive subscriptions with session timeouts.
opencodeGreat Swiss army knife of coding, packed with providers, agents, and integrations, if you want to embrace the complexity.
pi coding agentLighter than opencode, if you like customization with extensions, skills, and themes.
codehamrSimplicity over complexity, if you'd rather code than configure. An always-ready hamr that checks its own work, not just claims it's done.
free, DIY, or curated
Three ways to run.
We love local LLMs and will always stay MIT open source. You can also bring your own key with OpenRouter or OpenAI, chase the latest model, tune the parameters yourself. Or grab the optional HamrPass, outsource the whole thing, get a well crafted setup with the latest models tuned for codehamr, and support the project along the way.
Bring your own key
If it speaks OpenAI, codehamr speaks back. OpenRouter, your own proxy, your models, your bill. Flexible where it counts, but we avoid proprietary endpoints that lock you in.
HamrPass
Prepaid. No login, no auto-renew, no subscription. Burn tokens when you need them. We do the model hunt for you. Optimized for codehamr, ready out of the box.
Help codehamr grow. Tell us your hardware, grab a seat.
Ships when it's done.
for teams
Enterprise.
Two strengths. Both rare.
Budget, not seats. Tokens that last.
You buy a token budget for your whole team, valid until used up. Any team size, no per seat fees, no annual renewals. One pool, the whole team draws from it.
We pick the model lineup with care and tune it together with the codehamr client, so the LLM and the agent work as one rather than side by side. Your token budget stretches noticeably further, and it finally feels fair.
German hardware. German law.
Our own servers, located in Germany. Hardware we own and tune for codehamr, not a managed slice of AWS, GCP or Azure. Independent of hyperscaler outages and quotas.
German law is among the strictest data protection regimes in the world, and we operate fully under it. Your data stays in EU jurisdiction, never leaves our hardware, never gets used to train any model. Your DPO gets the documentation they need.
Tell us what your team needs, we'll size it together.
hardware
Run it on your machine.
Local LLMs finally caught up, and we love it. For the best experience, pick the largest local model that fits your RAM or VRAM — fully local and a real alternative to expensive cloud subscriptions. Less RAM on hand? HamrPass gives you an even better experience.
Info for Ollama users: Ollama Desktop may silently cut context to 4k. Open settings and lift the Context length slider to 64k+, depending on your RAM or VRAM.
install
Get it running.
Before you start, an honest AI disclaimer most tools don't tell you. codehamr is an AI agent that runs shell commands written by the model. AI sometimes gets it wrong, so run it in a sandboxed environment like a devcontainer or WSL2, and take a look at what it does. Take care, you're in charge.
curl -fsSL https://codehamr.com/install.sh | bashcurl -fsSL https://codehamr.com/install.cmd -o install.cmd && install.cmdOn first start, codehamr creates a .codehamr/config.yaml in your working directory. Edit to add more models.
.codehamr/config.yaml
# codehamr configuration
#
# Running codehamr on the same machine as Ollama? Keep localhost below.
# In a devcontainer/WSL2 with Ollama on the host? Start Ollama with
# OLLAMA_HOST=0.0.0.0, then use http://host.docker.internal:11434 (see the guides).
active: local
models:
local:
llm: qwen3.6:27b
url: http://localhost:11434
key: ""
context_size: 256000
openai:
llm: gpt-5.5
url: https://api.openai.com
key: sk-...
context_size: 128000
hamrpass:
llm: hamrpass
url: https://codehamr.com
key: hp_...learn more
Guides.
Quickstart: Linux / macOS
Install Ollama, pull a local model, point codehamr at localhost. No key, no cloud.
Quickstart: Windows 11
The shortest path. Ollama Desktop on Windows, codehamr in CMD, no Linux, no Docker.
Sandbox: WSL2 on Windows 11
Ollama Desktop on the Windows host, codehamr inside WSL2. The model uses your GPU, the agent can't reach your Windows files.
Sandbox: VS Code devcontainer
Codehamr inside a disposable Linux container, Ollama on your host. Two files, one config, works the same on macOS, Linux, and Windows.
Session: keys and commands
Paste in your HamrPass, learn the three slash commands, know which keys cancel a request.
Providers: OpenRouter, OpenAI, anything compatible
One wire format, one config block per provider. Switch with one line. Ollama, OpenRouter, OpenAI, HamrPass, all the same.
Concept: prove it, don't promise it
Why codehamr makes the agent check its own work as a habit, not a gate, and why it stays minimal so a local model has room to think.
your turn
It's hamrtime!
That's the whole thing. Simplicity over complexity, just a nice code hamr. Made for folks who'd rather code than configure. Swing it if you want, star it if you like, or hold a HamrPass seat.