An agent for everyone on your team.
Each person gets their own agent. It learns their work, acts with their accounts, and keeps going after they log off. You set the rules once, for all of them.
The whole team,
working while it sleeps.
Every agent works for one person. They can hand work to each other, and every one of them follows the same team rules.
Read 14 new preprints on her watchlist. Flagged three.
DONERerunning Figure 3 on the October dataset.
WORKINGDrafted the reply to reviewer 2. Sending needs his OK.
AWAITING APPROVALBooked cluster time for Thursday's runs.
DONEWaiting for the lab's data upload.
WAITING · NO COMPUTE CHARGEPrepared notes from yesterday's group meeting.
READY TO REVIEWSame idea. Any team.
A research assistant per researcher.
- Watches the literature overnight
- Reruns analyses in its own sandbox
- Drafts reviews, replies, and write-ups
A chief of staff per rep.
- Prepares every call from the account's history
- Updates the CRM after the meeting
- Writes the follow-up for a quick approval
An analyst per operator.
- Chases approvals and missing documents
- Reconciles reports across systems
- Flags what changed since yesterday
An executive assistant per lead.
- Triages the inbox before the day starts
- Pulls updates from the team's agents
- Drafts the weekly and board updates
Private where it should be.
Shared where it helps.
Their memory.
What the agent learns about one person's work stays with that person.
Their accounts.
Each agent acts with its own person's credentials, scoped to them. Credentials stay outside the sandbox.
Their conversations.
Session history is isolated per person. Teammates see shared results, not the private thread.
Team knowledge.
Connect the tools your team works in. Every agent can access the knowledge you've chosen to share.
Handoffs.
Agents ask each other for what they need, instead of guessing at a teammate's work.
One set of rules.
Shared policies, approvals, and budget limits apply across the team.
Set the rules once.
Choose what agents can do, what needs approval, and where spending stops. Constal enforces those rules on every call.
- Read shared Notion and GitHub
- ALLOWED BY TEAM POLICY
- Send email outside the company
- THE PERSON APPROVES
- Spend above your threshold
- TEAM LEAD APPROVES
- Reach the agent's budget limit
- HARD STOP
- Delete protected data
- BLOCKED
Example team policy. You choose the permissions and limits.
Three steps to a team of agents.
Pick the team
Choose who gets an agent and set the team's policy, approval rules, and budgets.
Connect shared tools
Connect Slack, Notion, GitHub, Linear, and email, with access scoped to your team.
Everyone says hello
Each person connects their own accounts and messages their agent in Slack, iMessage, or email.
Same platform.
Same guarantees.
Every agent runs on Constal, with the same durability, isolation, and controls available to teams building on the platform.
Picks up where it left off.
Sessions survive crashes and redeploys. Work can pause for days and resume with its context intact.
A person before anything sensitive.
Gate actions behind an approval. Your policy is enforced on every call, every time.
Everyone's work kept separate.
Each customer or person has their own identity and session history. Credentials stay outside the sandbox.
Replay any run.
Every step is checkpointed and every call traced. See what the agent saw, decided, and did.
Spend inside hard limits.
Set a budget and approval rules. Agents transact within those limits, with every transaction on the ledger.
No compute charge while waiting.
An agent can wait on a customer, a webhook, or a person without keeping compute running.
Priced per person.
One agent per person, with their own memory and accounts. Shared governance for the whole team.
Let's scope your setup.
Talk to us about your team, tools, and expected usage. We'll put together a plan for your workflow.
DISCUSS PRICING