An agent for the boring half of mapping
Autonomous mapping
agents.
Give an agent a goal, a slice of your inventory, and a budget. It uploads files, runs jobs, proposes matches, triages mismatches, and chases coverage — while you sleep. 7-day undo on every action.
What an agent does
Four jobs your team is doing today that an agent can take off your plate.
Pulls inventory
Reads from an S3 bucket or a webhook you configure, uploads new files through the same pipeline as the UI, and kicks off mapping jobs.
Runs and watches jobs
Triggers mapping jobs on a schedule or when new inventory lands, polls until they finish, and summarises results in plain English.
Proposes matches for unmatched rows
Searches the golden dataset by name and geo, files proposed matches as pending items for your review. It never auto-links — that stays your call.
Triages mismatches and chases coverage
Moves new mismatch reports into progress, groups duplicates, posts proposed resolutions, and pings you when coverage drops below your threshold.
Trust by design.
Autonomy is only useful when you can sleep through it. Three guarantees the system enforces, not just promises.
7-day undo
Every autonomous mutating action — mismatch transitions, mode-preference changes, pending-item creations — captures a snapshot before it runs. Roll any of them back from the Undo center within 7 days, even after the agent has moved on.
Hard cost caps
You set a daily LLM spend cap per agent. The cost governor checks the running ledger before every LLM call and pauses the agent at the cap. Approaching 80%? You get a notification. Nothing surprise-bills.
Human-in-the-loop by default
An agent never auto-links a result to a reference hotel, never marks a mismatch fixed, never pushes results to a sink unless you explicitly opt in. The defaults file pending items for your review; you stay in the driver's seat.
One page. Three decisions.
Identity and goal, scope and schedule, rules and limits. Most teams have their first agent live in under two minutes.
See a cycle, end-to-end
90 seconds. Inventory upload triggers the agent, it runs a mapping job, files two proposed matches, and posts a summary back to the thread.
No autoplay. No sound by default. Captions in the player.
Bring your own LLM key (optional)
Use the hosted Anthropic key included with Pro Max, or plug in your own Anthropic or OpenAI key per agent. Keys are stored in the same secret vault as your data-source credentials, redacted in audit logs, and only ever visible to workspace owners and admins.
FAQ
Ready to hand off the boring half?
Agents are included with Pro Max. Pick a plan, create your first agent, watch it work.