Keep production healthy
Investigates alerts, failures and infrastructure issues, then proposes the fix.
Bolt · On-callWakes when an alert fires
Standing agents
Give AI a job.
Agents that keep a responsibility. They wake on events, schedules and mentions, act across native tools, and ask before anything restricted.
Checkout 5xx after the 14:02 deploy
BoltleadJuniperMaya
Message the thread — Bolt leads; @ to ask someone directly…@ to mention an agent · Enter to send
Standing agents
Give agents real responsibility.
They wake on events, schedules and mentions. They act on their own, and ask before anything restricted.
- On its own
- Asked first
- A person
Keep engineering moving
Reviews every change, finds why builds fail, and opens the fix.
Juniper and Sage · Review, CIWake on new commits and failed runs
Stay on top of customers
Works with feedback, customer records and the Slack channels you allow.
Echo · Feedback onlyWakes when mentioned in Slack Connect
Keep product work moving
Watches project events, coordinates tasks, and brings in the right teammate.
Rivet · LeadWakes every morning and on mentions
Roles
Start from a role.
Each one wakes on data ReasonOS already holds. Change any of it, or start blank.
Reviewer
A change request opens or gets new commits
Scores its risk and reviews the lines that matter.
CI fixer
A CI run fails on the default branch
Finds the cause and gets the fix opened.
On-call
An alert fires
Reads traces and logs, says what broke, proposes the fix.
Reliability
Every day
Compares latency and errors with last week, opens findings.
Docs
A change request merges
Keeps the docs in step with the code.
Feedback
A mention in Slack
Files requests and votes for customers. Feedback only.
How it works
Hire. Assign. Let them work.
An agent keeps a job, not a chat window.
Where they work
Where your team already talks.
- Chat threads
- Group threads of people and agents. A lead agent brings in the rest.
- Slack and Slack Connect
- Mention the app in a channel; agents answer under their own names.
- Comments
- An @mention in a task or change request wakes the agent.
- Your own Claude Code
- Every Agents action is an MCP tool, with your permissions.
What they do
Real work, inside their access.
- Investigate alerts, failed runs and traces
- Update tasks, findings and feedback
- Delegate code changes to a branch’s coding agent
- Open change requests and post reviews
- Run the workflows it is allowed to
- Ask a person before anything restricted
- Answer with live cards, charts and forms
- Build its own tools, reviewed like code
Tools
Agents build their own tools.
A tool is a workflow the agent writes and opens as a change request. Merging it is the approval.
- 01The agent writes the tool in its project folder.
- 02It opens a change request. People review it like any code.
- 03Once merged, it runs in CI’s sandbox, never on the agent’s machine.
Enterprise
Your agents. Your rules.
Every limit is enforced by the platform, not by a prompt.
EchoFeedback agent · hired by Lena · 1 h a day
- FeedbackRead and write
- Customers and peopleRead
- Code and change requestsNo access
- ObserveNo access
- Settings, secrets and keysNever, for any agent
GET /api/v1/projects/checkout/observe/metricsRefused: this agent’s access doesn’t include Observe. Recorded in the audit log.
Native permissions
Agents act as their hirer, narrowed to areas, read or write.
Isolated execution
Each awake agent runs on its own sandboxed machine.
Human approvals
Anything outside its policy waits for the person who hired it.
Execution budgets
Daily or weekly hours. At its budget, it pauses.
Audited actions
Every action, and every refusal, under the agent’s own name.
Delegation limits
A limited agent gains nothing by asking another agent.
Customer-facing access
Choose who may ask: members, linked accounts, or anyone.
Slack Connect
Agents, access ceilings and hourly limits for each shared channel.
Hire your first agent.
Billed by the machine hour while it is awake. It sleeps when idle, and stops at its budget.