recuriant

The agentic operating system

Every agent action,
on the record.

For engineering and platform teams putting AI agents into production. Recuriant runs, connects and governs every agent action — inside Recuriant or from your terminal.

Start free Read the docs

one command, from your terminal

$ claude mcp add recuriant -- npx recuriant mcp
✓ recuriant added (streamable http, OAuth)
$ claude
> spin up a governed agent to triage today's alerts
→ dispatching via recuriant.run_agent(role: "sre-triage")
→ tier: L1 · running · decision logged

Recuriant runs RobotMind itself and the Actborne companies, end to end.

Four pillars

One system for the whole agent lifecycle

Run

Four agent runtimes — Claude Code, OpenCode, Hermes, OpenClaw — with capability-tier model routing across providers, durable Temporal execution, and a YAML workflow DSL with on/if/do/dag/verify/finally.

4 runtimes · durable execution · outcome verification

Connect

21 connectors and about 147 actions — GitHub, Stripe, chat, MQTT, ROS, Home Assistant and more — plus bring-your-own MCP servers, plus Recuriant itself exposed as an MCP server usable from Claude Code or Cursor in your terminal.

21 connectors · ~147 actions · BYO MCP

Govern

A policy enforcement point on every step, fail-closed. L1/L2/L3 tiers by reversibility and blast radius. L2 actions wait on a durable human approval. Every dispatch gets a short-lived, role-scoped identity whose permissions only narrow.

fail-closed PEP · L1/L2/L3 tiers · durable approvals

Recurse

Agents author and modify workflows, spawn other agents, and fire events — bounded by a depth limit, a concurrency cap, and scopes that only ever narrow.

bounded depth · bounded concurrency · narrowing scope

The signature idea

Recuriant is recursive

An agent in your terminal calls Recuriant over MCP. Recuriant runs a governed workflow. That workflow spawns another agent with a narrower, short-lived identity. A high-impact step pauses for a durable approval. Every hop is on the record.

01 Terminal agent
02 Recuriant MCP
03 Governed workflow
04 Spawned agent
05 L2 approval
06 Decision log

What it looks like in production

Real workflow. Real gate. Real record.

routes.d/deploy-hotfix.yaml

on: issue.labeled
if: "{label} == 'hotfix'"
do:
  - agent:
      type: code
      runtime: claude-helm
      role: sre-hotfix
  - stripe.refund:
      amount: "{incident.credit}"
      # reversible + high blast radius ⇒ tier: L2
verify:
  on: deploy.healthy
  within: 30m
finally:
  - matrix.send: "[{_status}] hotfix {issue.id}"

decision log

agent.dispatch role=sre-hotfixL1
stripe.refund amount=$412.00L2 · pending
approval.granted [email protected]signal
stripe.refund executedL2 · done
verify deploy.healthy within 30mpassed
finally matrix.senddone

Three ways in

Developers

Add Recuriant as an MCP server to Claude Code or Cursor with one command. Open-source core: SDK, MCP and workflow spec are Apache-2.0.

See the quickstart

Platform & security teams

Govern the agents you already run. Cloud or self-hosted / VPC, SSO, audit export, policy packs.

Read the governance model

SMBs

Recuriant Business Engineering: a done-with-you service where our hybrid team of humans and AI agents digitalizes your company on Recuriant, in fixed-scope sprints.

See Business Engineering

Pricing

Start free. Pay when you scale.

Open-source core, a free cloud tier, and paid plans for teams that need SSO, audit retention and self-hosting.

See full pricing

Put a governed agent to work today.

Start free, or add Recuriant as an MCP server to the terminal you already use.

Start free Read the docs