Tools for agents that need to act in the real world.
A developer-native layer for agentic workflows that need trusted real-world actions. Request a site verification, a measurement, an item retrieval, or a key handoff — with the same predictability as calling an API.
# Agentic workflow → trusted real-world action
from wishfleet import Wishfleet
wf = Wishfleet()
task = wf.verify_site(
address="1240 Broadway, Unit 4B",
evidence=["occupancy_photo", "lockbox_present"],
window="2026-04-22T09:00/12:00",
instructions="Confirm unit is vacant and ready for showing.",
)
result = task.await_result()
# -> { status: "completed", evidence: [ ... ], provider: "taskrabbit" }“Send someone to take the photos.”
Agents are moving from answering questions to completing workflows. Most of the tools they call are still digital — until a workflow hits the physical boundary. That’s where things stall today.
The moment it breaks
A property-ops agent needs someone to confirm a unit is vacant. A claims agent needs six photos of roof damage. A retail workflow needs a same-day measurement before issuing a quote. The next step isn’t a database write — it’s someone going somewhere.
What teams do today
Ad-hoc contractors, manual escalation, side-channel Slack threads, and glue code around marketplace booking APIs designed for human checkout. It works just enough to be painful, and the physical step never looks like part of the workflow.
What Wishfleet changes
Real-world work becomes a programmable capability. A defined schema, clear acceptance criteria, validation rules, and structured evidence on the way back — so the agent can keep reasoning and acting.
A library of real-world tools.
Named capabilities with predictable schemas. Each call includes instructions, constraints, and acceptance criteria — and returns structured evidence.
-
verify_siteConfirm a location matches the described state. Occupancy, presence of a lockbox, install status.
-
capture_photosStructured visual evidence against a shot list. Angles, subjects, and lighting are part of the contract.
-
measure_roomOn-site measurements returned as typed values, not a free-form note.
-
handoff_keysCoordinated physical handoff with a verified receiver, timestamp, and photo evidence.
-
retrieve_itemPick up, inspect, and deliver an item with condition evidence captured at each step.
Reliability, not escalation.
Wishfleet is designed to feel frictionless for the systems that call it — the developer, their framework, and the AI agent itself.
REST API
Clean keys, typed responses, a predictable event stream. Works the way you want it to.
CLI & SDKs
Organized by workflow types that you understand. Schemas and methods that LLMs get right the first time.