Autonomous agents, with guardrails you set
Agents can now start themselves when work arrives, follow instructions your team keeps in Hourtick, and stay inside the projects and budget you set.
What's new
- Webhooks start an agent when it has work: your endpoint, or a GitHub Action running Claude Code or Codex
- Instructions live in Hourtick, with every version, and lead every session
- Scope agents to clients and projects, or make them read-only
- Monthly budgets for model cost and time pause an agent and tell you
- Temporary agents expire on the date you pick
- Anyone can ask for an agent; admins approve on web or phone
- Connect any app as an agent right on the sign-in screen
- An activity log for every agent
An agent used to need a loop running somewhere, waiting for work. Now it doesn't have to. Give it a webhook and Hourtick calls it the moment someone hands the agent a task or mentions it, signed so you know it's us. Point it at GitHub's repository_dispatch and a workflow starts Claude Code or Codex headless, takes the work, logs its time and cost, and ends. No server, no idle cost.
What the agent should do lives in Hourtick too. Its instructions (how your team wants the work done) are kept with every version and lead the context of every session, so every app running it follows the same playbook. Agents can't change their own instructions.
More autonomy needs firmer limits, so admins now decide each agent's reach: the clients and projects it may work on (it doesn't see anything else), read-only, a monthly budget for model cost and logged time that pauses it and notifies you, and an expiry date for temporary agents. Members and agents can ask for a new agent, admins approve with one tap, and every change, connection and pause is in the agent's activity log. When you connect Claude, ChatGPT, Codex or Cursor, you can choose right on the sign-in screen to run it as one of your agents.
More in AI
Track your first hour in a minute.
Free for your whole team, forever. $29/month when you need 5 GB.