Autonomous Agents#
Most AI tools are reactive — you ask, they answer. Autonomous agents in LIT Platform work the other way: they run on a schedule, monitor things, and reach out to you when something needs attention.
This is the heartbeat feature. Your agent gets a pulse.
What Autonomous Agents Do#
An autonomous agent wakes up on a configurable interval, runs its assigned work, and sends you results through the channels and DM system. Between cycles it sleeps, consuming no tokens.
Example uses in data science workflows:
- Monitor a data pipeline and alert when a drift metric exceeds a threshold
- Run a nightly model evaluation and post results to a project channel
- Check an API for new data and summarize what arrived
- Review experiment logs and flag runs that need attention
- Send a morning briefing on the state of ongoing training jobs
The agent uses whatever tools it has configured — file access, web search, API calls via MCP — the same tools available in regular chat sessions.
Enabling Heartbeat#
Click the status dot next to any agent's name in the DM sidebar. A dropdown appears with four states:
| State | Behavior |
|---|---|
| Disabled | Agent only responds when you message it |
| Enabled | Agent wakes on its configured schedule and runs autonomously |
| Safe mode | Agent runs autonomously but pauses for confirmation before taking any action that modifies state |
| Stopped | Agent is paused mid-cycle; resumes from where it left off |
The agent runs as long as the platform is running. On restart, heartbeat agents resume automatically.
The Inner Monologue is the Output#
When a heartbeat agent runs, you see everything — tool calls, intermediate reasoning, the full stream — posted to the channel in real time. There's no separate "summary" layer. The agent's natural response is the deliverable.
This keeps the system simple and auditable. You can read back through any cycle's work just by scrolling the channel history.
Reaching Out to the Agent#
Heartbeat works both ways. You can message the agent in its DM thread at any time — mid-cycle or between cycles — and it will respond in the same channel. The persistent channel history means the agent has context on prior cycles when you ask follow-up questions.
A Colleague, Not a Cron Job#
The difference between a scheduled script and an autonomous agent is judgment. A cron job runs the same code every time. An autonomous agent adapts — it decides what's worth reporting, asks clarifying questions when something is ambiguous, and learns the shape of normal vs. anomalous over repeated cycles.
This is what makes heartbeat mode useful for data science work specifically: your data doesn't behave the same way every day, and neither should your monitoring.
Team Deployments#
On multi-user LIT Platform deployments, heartbeat agents can post to shared team channels. Multiple team members receive the same updates and can reply in the same thread. The agent maintains context across all participants.