Praecise publishes the tools it runs on.
Four public repositories: the agent framework, the inference layer, its pinned backend and the open operating system base. Each is described here as its own README describes it.
In short
Praecise publishes Harness, Engine, a pinned llama.cpp backend and Furcate in public repositories. It is for developers who want to read, run or build on them. Each repository's README has the start commands.
00
At a glance.
Harness
Apache-2.0.
npm @praecise/harness, command praecise, Node 22 or newer.
Engine
Apache-2.0 and MIT.
SPDX Apache-2.0 AND MIT, as its README states.
Furcate
Apache-2.0.
github.com/furcateresearch/furcate.
01
The repositories.
- 01
The framework for AI agents. A folder is an app. TypeScript, Node 22 or newer, npm package @praecise/harness, command praecise.
- 02
A backend-agnostic inference-acceleration layer for large language models. llama.cpp is the first backend. Not a model server on its own.
- 03
The pinned llama.cpp backend fork that Engine builds against. © the ggml authors.
- 04
The open base for private, owner-controlled local AI: OS base, identity, signed updates, leaf firmware base.
02
How to start.
Each project starts from its README.
# Harness npx @praecise/harness init my-app cd my-app && npm install npx @praecise/harness dev
Engine is consumed as a Rust crate: depend on praecise-runtime, which defaults to no backend. See its README, section Consuming Praecise Engine.
03
What each is for.
- 01
Harness
Make an agent work.
Models, escalation, grounding, memory and resumable workflows, served as chat, REST, MCP and A2A.
- 02
Engine
Make a model faster.
Speculation policy, drafter selection, request shaping and admission. Neither Harness nor Engine needs the other.
- 03
Measured, with caveats
Structured text only.
Engine is measured on structured text such as SQL and code, and shows no gain on prose. The Engine README has the method and the limits.
- 04
A bandwidth ceiling
Stated plainly.
On a DGX Spark, decode runs near 80 percent of the 273 GB/s peak, per the Engine documents.
- 05
Furcate
Run your own AI on your machines.
One headless Linux system on every machine. Furcate is Linux only.
- 06
llama.cpp fork
The backend Engine uses first.
Kept under its own MIT licence.
04
Where each runs on Cloud.
Harness
Placed as a Node app.
With one key it uses Praecise Cloud's models, database and object storage.
Engine
Accelerating models on Metal.
Engine accelerates the models served on Praecise Metal.
Furcate
Under everything.
Praecise Cloud runs on Furcate, and so can your machines.
05
Open source and the products around it.
Praecise Engine
The acceleration layer under a model, so the same app gets faster models on Praecise Metal.
Furcate
The open operating system base: hardware-rooted identity and signed updates.
Praecise Cloud
Build, Place and Operate apps, agents, models and data.
Praecise Metal
Servers and GPU nodes that Praecise operates, dedicated or managed.
Praecise Silo
An isolated computer for every agent, which reaches only the services you connect to it.
06
Read before you rely on it.
- 01
Harness persists in plaintext
Runs are sealed at rest only with a run cipher.
- 02
Furcate status
The README separates what runs on hardware, what is tested in software and what is in progress.
- 03
Engine does not claim a general speed-up
Gains are measured on structured text and absent on prose.
- 04
Private repositories
Furcate Enterprise, firmware and hardware repositories are private and not described as open.
- 05
npm scope
Only @praecise/harness is published under it.
- 06
Contributions
Open an issue or a pull request on the repository.
07
Questions and answers.
- 01
What does Praecise publish as open source?
Praecise Harness, Praecise Engine, a pinned llama.cpp backend fork and Furcate, each in a public repository.
- 02
What are the licences?
Harness and Furcate are Apache-2.0. Engine is Apache-2.0 and MIT. The llama.cpp fork stays MIT.
- 03
How do I start with Harness?
Run npx @praecise/harness init my-app, then npm install and npx @praecise/harness dev. It needs Node 22 or newer.
- 04
Is Engine a model server?
No. Engine is an acceleration layer between an application and an inference backend. llama.cpp is the first backend.
- 05
Which operating systems does Furcate run on?
Furcate is Linux only, per its README.
08
Next pages.
09
Praecise accounts are by invitation. Open the console, or talk to us.
New to Praecise Cloud? Accounts are set up with the Praecise team: talk to us, then sign in to the Cloud section of Praecise Spaces with an email code.