Build agents. Or let the platform build them for you.
Tokoaido is not a black box that either does everything or nothing. Author your own agents with a no-code blueprint, run the built-in fleet out of the box, and let the platform recommend and build new agents from what it discovers. Every one draws on a governed tool library and earns its autonomy.
Bring your expertise, or borrow ours.
You author them
Describe an agent as a blueprint — a signal to watch, a finding to flag, a governed action to propose. No code. It compiles to a real, sandboxed agent on the same governed surface the platform uses.
The platform ships a fleet
A tiered fleet runs out of the box — sensing agents watch the streams, diagnostic agents find root cause, optimizers propose actions, an orchestrator resolves trade-offs and dispatches under governance.
The platform proposes them
It discovers your real processes, recommends automations ranked by ROI and risk, generates connectors from a vendor's docs, and improves its own agents over time from measured outcomes.
From a signal to a governed agent in minutes.
A domain expert — a reliability engineer, an ops lead — describes an agent as a blueprint. No prompts to wrangle, no pipeline to wire. It compiles to a real, sandboxed agent that runs on exactly the same governed surface as the platform's own.
- 01Choose a signalA machine tag or a system event to watch — a pump's vibration, an SLA field on a ticket, a spend threshold on a PO.
- 02Set the conditionRises above, falls below, or trends — with a threshold or a window. Plain choices, no query language.
- 03Describe the findingWhat it means, in plain language, with live values filled in — so a human reviewer immediately understands.
- 04Define a governed actionA write to an IT system through a connector, or an OT change routed through MES/SCADA. The Safety Instrumented System and direct-PLC targets aren't even expressible.
- 05Dry-run, then publishTest the agent against live telemetry before it goes anywhere. Publish it at advisory, where it earns autonomy from its track record.
A builder that can't build something dangerous.
A raw agent builder lets you write a prompt and hope. Ours makes governance the substrate — so the agents you build are the ones you can actually put into production.
New agents start advisory and are promoted only by a measured track record — never autonomous on day one.
Consequential actions wait for a person at the right risk tier; nothing high-risk executes silently.
A direct PLC write or any Safety Instrumented System interaction is blocked in the architecture — unexpressible, not just discouraged.
Every agent is evaluated continuously against golden scenarios, and every action carries tamper-evident lineage.
An agent is only as capable as its tools.
A tool is any governed capability an agent calls to gather, reason, and act: read a PDF, run a three-way match, query a system of record, adjust a setpoint through MES. The library ships with 128 tools across 11 categories, and it keeps growing.
Read and understand any document: template-free extraction, OCR, and grounded answers with a citation behind every claim.
Point the connector-from-docs agent at an API's documentation and it generates a working, governed connector.
The live HTTP and OAuth connector turns any web API into a tool an agent can use.
Open A2A and MCP interop lets external tools and agents participate under the same governance.
The point: any system you can connect becomes a tool, and the platform equips the agents it builds from exactly this library. So agents can take on virtually any task across your operation, always within the governance and the hard safety boundary.
Build your first agent in the live workspace.
Author a governed agent, dry-run it against live telemetry, and publish it at advisory — no signup.