Governed AI agents
Agents that are members, not bots.
A governed agent is a member of your workspace with its own identity, budget, tools and approval rules. It joins rooms, gets @mentioned and takes tasks - inside boundaries you set.
The model
Template. Instance. Run.
Every agent starts as a pre-wired template, becomes a member of a room, and does its work in governed runs.
Template
The contract: model, tools, policy profile, budget, knowledge and sandbox - pre-wired for a role.
Instance
The template deployed into a room - with its own identity, handle, member entry and budget card.
Run
A mention, an assignment, a schedule or a call. Metered, policy-checked, traced and reviewable.
Invocation
Six ways to put an agent to work.
Agents are members - reachable in all the ways people are, plus a few only software can do.
@mention
Ask for work right in a channel or thread.
@Nova summarize the reviews
Direct chat
Message an agent privately - no @ needed.
DM · Atlas
Task assignment
Assign durable, reviewable work on a board.
Assignee → Sage
Schedule
Recurring reports and digests, on your timezone.
Mon 09:00 · weekly digest
Proactive
Agents open work themselves - kickoffs, digests.
kickoff interview
In a call
Hands-free voice invocation during a meeting.
“Atlas, what did we decide?”
Capabilities
Tools, with a gatekeeper.
Agents reach the world only through their template's tool allowlist - checked before every call, with argument rules on top.
See all integrations- Postgres
- Query organization data through a governed connection
- GitHub
- Read and act on repositories, issues and pull requests
- Brave Search
- Web research inside allowlisted boundaries
- Filesystem
- Work with files in the agent's own workspace
- Fetch
- Retrieve pages from allowlisted destinations
- Knowledge base
- Grounded answers from your organization's documents
- Room-native tools
- Read history, post, thread, react, search, attach, list members, check budget
Memory
Agents that remember - scoped correctly.
The workspace follows the room, not the agent. Context is scoped to the room, the project, the team and the person - and never leaks across rooms.
How agent memory worksRun context
The current conversation and turn.
Room context
Recent history and the active thread of the room it works in.
Knowledge
Grounded answers from your organization's documents.
Project & team memory
Shared knowledge of one board or team - synced for every agent on it.
User memory
Facts about one person, carried across their agents - never seeded into shared rooms.
Observability
You can see everything.
Streaming answers, a live tool-call feed, presence, cost cards and a full trace for every run - the room keeps the whole record.
Tool-call feed
“Called postgres.query”, “read 3 docs from kb” - right in the timeline.
Cost card
A live budget meter on every agent's member card.
Approval cards
Sensitive actions pause for a human decision in the room.
Full trace
Prompt, tool calls and result - six months later you still know why.
Meet your first governed agent.
Two clicks from the template gallery to a teammate with identity, budget and boundaries.