Put your AI agents on the team
Add your own AI agent to Manifestly as a team member. Assign it steps, invite it to runs, and @mention it, the same way you would a person. It does its part of the workflow and your people do theirs. Every action is permissioned and on the record.
AI Agents is in beta and turned on per account. Talk to support to try it.
- Collect W-9 and banking details DR
- Verify tax ID and screen vendor AI agent
- Create vendor record in ERP AI agent
- Approve vendor Approval PS
Agents need a place in the process, not just access to it
An agent with an API key can do things, but nobody knows what it's responsible for, when it should act, or what it did. AI Agents gives your agent a seat on the team: assigned work, a defined scope, and its own line in the audit trail.
- Acts as whoever's key it was given
- Has to poll to find out what changed
- Its work is mixed in with a person's activity
- Its own identity, department, and role
- Notified the moment work is assigned to it
- Every action attributed to the agent by name
From new hire to working teammate in three steps
Add the agent
An account admin adds an AI agent to a department and gives it a name, a role, and the webhook URL where its work should go. Manifestly issues the agent its own API key.
Hand it work
The agent shows up in the same pickers as your people. Assign it steps, add it to a role, invite it to runs and scheduled runs, or @mention it in a comment.
It does its part
Manifestly sends a signed webhook the moment something needs the agent. The agent acts back through the Manifestly MCP server or the v2 API with its own key. Its work lands in the run like anyone else's.
Your agent hears about work the moment it's ready
Your people get email, Slack, and Teams notifications. Your agent gets a webhook. When a step is assigned to it, becomes ready, is mentioned in a comment, or runs late, Manifestly posts a signed event to the agent's endpoint right away.
run_invited
step_assigned
step_became_applicable
step_role_ready
mention
comment_created
step_completed
approval_rejected
step_late
run_almost_late
run_late
run_completed
reminder
- Each delivery is HMAC-signed with your account's webhook secret.
- Payloads carry IDs only, never run content. The agent fetches the details it's allowed to see with its own key.
- Failed deliveries retry automatically. Admins can see every delivery and retry it by hand.
# Manifestly → your agent's endpoint
POST https://agents.example.com/manifestly
X-Manifestly-Event: step_assigned
X-Manifestly-Delivery: 8f2c…e41a
{
"event": "step_assigned",
"occurred_at": "2026-10-12T14:03:22Z",
"agent_id": 4182,
"run_id": 90231,
"run_step_id": 771045
}
# → the agent reads the step over MCP or the v2 API,
# does the work, and completes it
Every action has a name on it
When the agent completes a step, records data, or leaves a comment, the activity feed shows the agent as the one who did it. It isn't credited to the admin who set it up or to a shared integration key. When someone asks who verified a vendor, the record answers.
- Agent work is attributed to the agent in run activity and reporting.
- Agents are clearly marked with a robot avatar wherever they appear.
- Admin actions on an agent, such as rotating its key, are logged too.
-
PS
Priya Shah approved Approve vendor2:41 PM
-
Vendor Setup Agent completed Create vendor record in ERP2:12 PM
-
Vendor Setup Agent commented: “Tax ID verified. No sanctions matches.”2:09 PM
-
DR
Dana Ruiz completed Collect W-9 and banking details1:55 PM
Bring the agent you already run
Manifestly doesn't host your agent or choose its model. Use whatever you've built or bought: a custom LLM agent, an agent framework, or an automation platform. Manifestly stays the system of record for who owns which step, what state the run is in, what's allowed, and what happened.
- Act through the Manifestly MCP server or the v2 API.
- Each agent authenticates with its own API key, shown once at creation and rotatable at any time.
Autonomous, but never unaccountable
- Scoped like a person. An agent sees and changes only what its department and role allow. An agent in Finance can't read HR.
- Humans keep the final say. Put an approval step after the agent's work. If a reviewer rejects it, the agent is told so it can rework it.
- Admin-controlled. Only account admins can add agents, view their deliveries, or rotate their keys. Agents can never sign in to the app.
- An instant off switch. Rotate the key and the old one stops working immediately. Clear the webhook URL and deliveries stop.
| Sent | Event | Result |
|---|---|---|
| 2:41 PM | run_completed | Delivered |
| 2:12 PM | step_assigned | Delivered |
| 2:03 PM | step_became_applicable | Delivered |
| 1:58 PM | mention | Delivered |
| 1:55 PM | run_invited | Delivered |
Where teams put an agent to work
Finance & procurement
Vendor checks, invoice matching, and system-of-record updates.
The agent verifies the tax ID and creates the ERP record. A controller approves before payment is set up.
IT & employee onboarding
Account provisioning and access requests across your tools.
HR completes the hire details. The agent is notified, creates the accounts, and completes its steps.
Compliance & reporting
Evidence gathering, drafting, and recurring status summaries.
Each month the agent pulls the evidence, attaches its findings, and @mentions the reviewer.
AI Agents, answered
Other Features
Workflow Customization
Shape exactly how each workflow behaves — logic, data, timing, and documentation.
Assign and Automate
Get the right work to the right people, on schedule — automatically.
Track Progress and Reporting
See what's done, what's late, and what's coming — across every run.
Govern, secure & prove it
SSO, granular permissions, and a complete audit trail — the controls enterprise teams need to run their work and prove how it was done.
Communicate
Keep the conversation — and a permanent record of who did what, and when — in one place.
Trusted by organizations big and small for their recurring workflows
Put Your AI Agents On The Team
With Manifestly, your team will Never Miss a Thing.