Open source · Apache-2.0 · v1.1.0
The framework for AI agents.
A folder is an app. You describe what an agent is for, what it knows and what it may act through. Which model to use, when to escalate to a stronger one, how to ground an answer in your files and how to resume a workflow that stopped for a person are handled for you.
npx @praecise/harness init my-appWrite the agent. The rest is derived.
One file per agent, in a folder. Running it gives you a chat page, a REST API and an MCP endpoint for that agent.
my-app/
├─ agents/support.ts
├─ functions/refund.ts
├─ memory/faq.md
└─ .envimport { agent } from "@praecise/harness";
export default agent({
role: "Customer support for Acme.",
tools: ["refund"],
});npx @praecise/harness dev
# → http://localhost:3000/support
# chat UI · REST · MCP endpointWhat is in the folder
Nothing is required except the one folder you use. Each of these is a file you write and a page of the documentation.
Agents
A role, what the agent knows and the tools it may use. There is no model id to choose or keep current.
Workflows
Steps that run in order, wait for a person where you say so, and resume from where they stopped.
Tools
Your own functions, and anything that speaks MCP, one file per service.
Stores
Memory and data an agent can answer from. praecise ingest reads PDF, Word, Excel, images and source into one.
Evals
Say what a right answer looks like, so a change to a role, a tool or a model is checked before it ships.
Plugins and environments
A package extends an app from one line of config. An environment equips agents for a whole platform.
Run it anywhere
A dashboard, chat, REST, MCP and traces in development. Package the app, or deploy it to Praecise Cloud.
What you are trusting
The server authenticates and refuses to start on an address the network can see without it. What is assumed is written down.
Your endpoints, or one key
Complete on its own with your models, database and storage. Praecise Cloud is the managed side of the same app.
The documentation
Every page is written in the project's own repository and built from it, so the site and the code say the same thing.
Building
Running
Around the harness
The harness runs agents. These sit beside it and none of them is needed for it to be useful.
Environments
A platform's skills, tools, agent and checks in one plugin. Apple development is the first.
Praecise Engine
The acceleration layer underneath a model: speculative decoding, batching and backend tuning.
Praecise Cloud
One key, and the models, the database and object storage are provisioned for the same app.