Cerveau logo

Cerveau

Free

Local-first agentic coding harness — Go + Svelte + llama.cpp, engineered for small local MoE models on consumer hardware. cerveau.sh

Visit Website

At a glance

Explore the listed capabilities, then check the details that matter to you.

  • Run coding agents on local hardware
  • Switch inference cores between tasks
  • Preserve context across inference core changes
Related by shared tags and keywordsdeepseek-harness-setupqwenpippa
ai-agentcoding-assistantgolangllama-cpplocal-llmaiai assistantautomation
Cerveau screenshot 1

Overview

Cerveau is a local-first agentic coding harness designed to run on the user's own hardware. It focuses on enabling small models to perform real work on consumer-grade GPUs, without relying on cloud APIs or incurring hidden costs. The system is built to exploit existing machines rather than requiring rented cloud resources.

The harness supports two inference cores: llama.cpp and vLLM. llama.cpp is the default core, which offloads mixture-of-experts weights to system RAM, while vLLM keeps dense model weights resident on the GPU for higher throughput. Users can switch between cores at task boundaries, with session context preserved in Typesense and the embedder.

Cerveau includes memory management that treats memory as state, with features like typed records, self-correcting memory, and demotion instead of summarization. Safety is compiled into the system as Go code at the dispatch point, blocking destructive commands before they reach a shell. The tool binds to localhost and operates without cloud, accounts, or telemetry.

Pricing

Free

Detailed plans have not been confirmed in our catalog. Check the official website for current limits and billing terms.

Visit Website

Prices and limits may change. Confirm the currency, billing period, seat minimum and usage caps on the official website.

Use Cases

  • Teams use Cerveau to run large mixture-of-experts models on a 24 GB card by offloading experts to system RAM.
  • Teams use Cerveau to run dense models with real throughput using vLLM.
  • Teams use Cerveau to access the agent remotely from an Android phone over a tailnet.
  • Teams use Cerveau to start builds before leaving and check tool calls later.
  • Teams use Cerveau to answer agent questions remotely and avoid delays.
  • Teams use Cerveau to keep code and data local for small businesses.

Features

Run coding agents on local hardware

Switch inference cores between tasks

Preserve context across inference core changes

Store structured, self-correcting agent memory

Block destructive commands before shell execution

Keep operations bound to localhost

Reviews

No reviews yet. Be the first to share your experience!