Everything an agent needs, from one stack.
The agent, the model and the machine sit under one account. Each agent can have its own computer, its own address and its own payment limits.
In short
The agent stack is nine layers that an agent uses: Harness to build it, Engine to speed up its model, Silo to run code, Inbox for an address and Skerry for payments. It is for people building agents who want them under one account. Start with the Harness folder, then place it.
00
At a glance.
Open source
Harness and Engine.
The agent framework and the inference layer are public.
In use
Praecise runs on them.
Harness, Engine, Cloud and Inbox are in production use at Praecise.
Operations
Rivier sits on top.
Rivier puts the agents to work across a business's tools.
01
The nine layers.
- 01
A folder is an app. Write what an agent knows and can do; models, escalation, grounding, memory and resumable workflows are handled. Every app is served as a chat interface, a REST API, an MCP endpoint and an A2A peer.
- 02
An inference-acceleration layer between the application and a backend. llama.cpp is the first backend. It adds speculation policy, hardware-aware tuning and one serving surface.
- 03
Models
Managed models through one key, or your own endpoints and keys described in the app's config. Access to models and providers as a service in its own right is planned and not available yet.
- 04
An isolated computer for every agent, which reaches only the services you connect to it.
- 05
Automations
Workflows that wait for events, stop for a person to approve and resume where they left off.
- 06
Connect
Agents reach other systems through MCP servers and OpenAPI documents turned into tools, behind a guard the app owns.
- 07
One account across Praecise products.
- 08
An address and a searchable record for every agent and person.
- 09
Payment rails for agents, orchestrated across licensed partners, inside limits you set.
02
What only this stack gives an agent.
- 01
A computer of its own
Agents kept apart.
A Silo is an isolated computer with its own disk, started for one agent.
- 02
An address of its own
It can be written to.
An Inbox address on your verified domain, with a searchable record of everything said.
- 03
A budget of its own
It pays inside limits.
A Skerry key per agent with spend limits, approvals and revocation.
- 04
A model beside it
Same account, same stack.
Managed models through one key, and Engine accelerating the models served on Praecise Metal.
- 05
A folder as the whole app
Everything is a file.
Agents, workflows, memory, functions and stores live in one folder that Harness serves.
- 06
Work that waits
A person can step in.
Workflows stop at an approve step and resume where they stopped.
03
Write an agent.
An agent is a role and the tools it may use. The app around it is derived.
// agents/support.ts
import { agent } from "@praecise/harness";
export default agent({
role: "Customer support for Acme.",
tools: ["refund"],
});npx @praecise/harness init my-app starts a new app; npx @praecise/harness dev runs it with a dashboard, chat, REST, MCP and traces.
04
Build, Place, Operate.
Build
Write it and ground it.
A role, tools, memory files and a guard. Doctor checks the folder in one pass.
Place
Give it a home.
Place the app behind the edge, then give each agent its Silo, Inbox address and Skerry key.
Operate
Watch and approve.
Traces show what a request did, and approve steps hold risky actions for a person.
05
The products behind the layers.
Praecise Engine
The acceleration layer under a model, so the same app gets faster models on Praecise Metal.
Praecise Silo
An isolated computer for every agent, which reaches only the services you connect to it.
Praecise Inbox
An address and a searchable record for every agent and person.
Skerry
Payment orchestration for agents across licensed partners, with scoped keys and spend limits.
06
Agents that fit.
- 01
A support agent with its own inbox
Answers from your documents and replies from its own address.
- 02
A booking agent
Takes bookings and settles through Skerry, in sandbox today.
- 03
A coding agent in a Silo
Runs commands with no network and a receipt for every run.
- 04
An operations agent
Rivier puts agents to work across the tools a business already uses.
- 05
A tool for other agents
Every Harness app is an MCP endpoint and an A2A peer.
- 06
A model-serving agent
Engine speeds up an open model on Metal.
07
Questions and answers.
- 01
What is the Praecise agent stack?
Nine products an agent uses: Harness, Engine, Models, Silo, Automations, Connect, Identity, Inbox and Skerry, under one account.
- 02
What is Praecise Harness?
The open framework for AI agents, where a folder is an app. It is served as a chat page, a REST API, an MCP endpoint and an A2A peer.
- 03
What is a Silo?
An isolated computer started for one agent, with its own disk. Its only ways out are the services its owner binds to it.
- 04
How does an agent get an address?
Praecise Inbox gives an agent a mailbox on your verified domain, with everything said kept in one searchable record.
- 05
How does an agent pay?
Through Skerry, with a key scoped to the agent, a spend limit, approvals and revocation. Agent payments are sandbox today.
- 06
Is Harness open source?
Yes. It is Apache-2.0 at github.com/Praecise/praecise-harness, published on npm as @praecise/harness.
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.