Domain Experts view
Build AI agents on your expertise.
The terms your business uses, the rules it follows, the steps it takes, the voice it speaks in. Your domain experts author them as durable, versioned, schema-enforced components the agent runs on, so even Co-Pilot or a coding agent can't create malformed context. Not prompts buried in code.
What it knows
How it behaves
Authored by domain experts · Versioned and reversible · Governed, not hard-coded · Composed fresh every turn
Authored
By domain experts, no code
Versioned
Every change reversible
Governed
One definition per agent
Composed
Assembled fresh per turn
Components
“Does it follow your rules and speak your language?”
Knowledge & search
“Can it find and ground its answers?”
Startup evaluation
“Is it right before it goes live?”
Manifest
“What ties it all together?”
The rules, words, and voice your agent runs on, authored as versioned components.
The procedures it follows, the vocabulary it speaks, the voice it uses, and the actions it can take. Your experts author them in plain language as durable, versioned, schema-checked components, so even Co-Pilot or a coding agent can't create malformed context. Not prompts buried in code.
Routines & Policies
The procedures the agent follows and the rules it applies. If this, then that, and the steps to take, written in plain language and applied every time.
Glossary & terms
The vocabulary your business runs on, and the other names people use for the same thing, so the agent understands you the way your team does.
Prompts & instructions
The voice it speaks in and the standing directions it follows between turns. Authored as objects your team owns, not pasted into code.
Tools
The actions the agent is allowed to take in your systems: issue the refund, look up the order. Connected once and scoped to each agent.
The knowledge behind the answers, searched the way your team asks.
Point the agent at your documents and it searches them for you, by meaning, by keyword, or both. Authored in Co-Pilot or the CLI and kept current by the people who own the material, so it grounds answers in your reference instead of guessing.
Search, built in
The agent searches your material by meaning, by exact term, or both, so it finds the right passage however the question is phrased.
Authored in the CLI or Co-Pilot
Add, connect, and update the agent's knowledge yourself in Co-Pilot, or with the CLI. No separate pipeline, no ticket.
Powered by Postgres
Built on Postgres, a proven, serious database, so your knowledge is fast to search and dependable in production.
Kept current by its owners
Refreshed by the people who know the material; the agent always grounds its answers in the latest set.
Search method
Checked before it goes live, so it's right in production.
Before an agent starts serving, its context is validated to be clear, consistent, and complete. Anything malformed is caught up front, so what reaches your customers behaves the way your experts intended, not discovered mid-conversation.
Validated
Every rule, term, and prompt is checked before the agent goes live, so nothing malformed reaches your customers.
Clear and optimized
Gaps and conflicts are surfaced early, and the context is tidied into the shape the agent runs on.
Builds the context store
Once it passes, the agent is ready to assemble the right context on every turn.
Fail fast, not in production
Problems show up up front with a clear message, not mid-conversation with a customer.
An agent isn't a prompt. It's a definition.
The Manifest is the agent's definition: which glossaries, policies, routines, prompts, and macros make it up, and which exact version of each. Versioned as one object, promoted as one object, rolled back as one object.
One governed object
What this agent knows and how it behaves, in a single definition you own.
Composed, not pre-stuffed
The runtime assembles only the objects each turn needs. Nothing over-loaded.
Versioned and reversible
Promote a change as one object; roll the whole agent back in one move.
Built from your objects
Every glossary, policy, routine, and prompt you authored, named by exact version.
Binds an exact version of each piece
Built together
The rules your business runs on, owned by the people who know them.
No ticket, no sprint, no translating requirements into code. Your domain experts author the context directly; your engineers extend it where it needs code. One source of truth, governed together.
Domain experts
Author the glossaries, policies, routines, and prompts, in plain language. No code, no ticket, no waiting on a sprint.
Author with Co-PilotEngineers
Extend with custom tools, integrations, and the model, as code, against the same versioned catalog.
Build with the CLI & SDKOne versioned catalog · One source of truth · Governed together
Context
Turn what your team knows into the agent itself.
Bring a workflow and the rules behind it. We'll map how your experts build it as context the agent runs on, every turn.