The platform

Production AI agents, built, governed, and observed in one place

Describe an operation, compile it into a visual playbook, reuse specialist skills, and deploy the same governed agent across the channels where work happens.

Build a playbook live View pricing

1,066

Catalog entries

212

Marked live

44,421

Actions

308

Triggers

What the platform does

Agent builder

Generate proposal-first visual playbooks from plain English, then refine branches, tools, and approvals in conversation.

Visual + code-first agent definitions

Stateful sub-agent delegation

Human-in-the-loop checkpoints

Versioned, diff-able agent specs

Silent Observer and reusable skills

Discover repeated work with local sensitive-data filtering, then package instructions, tools, files, and knowledge into reusable specialist capabilities.

Privacy-first routine discovery

Value estimates before automation

Always, automatic, or on-demand knowledge

Side-effect-blocked simulation

Multi-channel deployment

Take the same governed agent to web, API and webhooks, Slack, Discord, or WhatsApp without rebuilding its policy model.

Web widget

API and webhooks

Slack and Discord

Tool catalog

One catalog of 1000+ REST / MCP / database / GraphQL integrations with OAuth, BYO credentials, and per-tool policy.

Hosted OAuth or bring-your-own client

Automatic schema derivation from OpenAPI

Per-workspace credential isolation

Custom HTTP tools without code

Governance & policy

Approve, sandbox, and rate-limit agent behaviour at the platform level. The same policy is enforced wherever the agent runs.

Resource access groups & scope tags

PII redaction at the tool boundary

Mutation guardrails (read vs. write)

Audit log with full request lineage

Observability

Trace every prompt, tool call, retry, and token. Replay any agent run with the exact inputs, secrets redacted.

OpenTelemetry-compatible traces

Per-step latency, tokens, and cost

Trigger-based webhooks for incidents

Production-grade analytics + alerts

System Architecture

How it's built

Four runtime engines sit under one control plane. Choose the right execution surface for each workload while sharing one identity and policy model.

Control plane

Workspaces, identity (SSO + RBAC), credentials, billing, and the policy engine that every runtime obeys.

Agent runtime

Executes agent specs deterministically. Supports parallel sub-agents, retries, and human approval steps.

Tool runtime

REST, MCP (stdio + streamable HTTP), database, and GraphQL drivers are invoked through one auth-aware client.

Trigger runtime

Webhooks, schedules, and event sources kick off agents and capture every payload for replay.

A catalog of 1,066 integration entries

The current catalog snapshot contains 212 entries marked live. Availability and authentication method are shown per integration instead of treating every catalog entry as equally ready.