The control plane for OpenClaw fleets — and every other agent. Governance, security, compliance, and observability built in. Bring your own agents; Recursant gives them identity, policy enforcement, and an audit trail that satisfies a security review.
The problem
AI agents are spreading faster than we can govern them
Frameworks, tools, models, clouds — every team is shipping agents into production. The governance layer has not kept up.
The Solution
The control plane for AI agents
Sits between agents and the systems they talk to. Discovers, traces, and enforces — across any framework, cloud, or runtime.
Today: ungoverned agent sprawl0 agents · ungoverned
Tomorrow: every agent governed40 agents · governed
✕No discovery of ungoverned agents
✓Automatic discovery across every environment
✕No visibility of agent activity
✓Full visibility of every agent decision and action
✕No audit trail
✓Hash-chained, regulator-ready audit across stacks
✕No enforcement of policy
✓Real-time, inline policy enforcement on the network layer
Register every claw. Govern the fleet.
OpenClaw is the fastest-growing personal AI agent platform. Recursant is its control plane. Spin up 50 claws, load the plugin, and every agent appears in your registry — governed, audited, and policy-enforced from a single dashboard.
Fleet registration
Install the Recursant plugin into each OpenClaw gateway. Every agent auto-registers, goes through security review, and appears in your control plane — no manual wiring.
Governed lifecycle
Each claw progresses from DRAFT through automated security testing to ACTIVE. No ungoverned agent reaches production — every one is reviewed before it acts.
Uniform policy enforcement
PII redaction, compliance rules, rate limiting, and a tamper-evident audit trail — applied uniformly across every claw in the fleet from a single registry.
Full fleet visibility
Every tool call, every LLM prompt, every inter-agent message — traced and logged. Live mesh visualizer shows you exactly what your fleet is doing in real time.
Architecture
Mesh enforces agent policy in real-time
Recursant is built on the service mesh pattern with a control plane and a data plane, the same pattern as Istio for microservices.
Control plane
Registry
Flask + React app backed by PostgreSQL, Redis, and Kafka. The single source of truth for agent metadata, policies, certificates, and audit history. Includes a full web UI for governance workflows and a live mesh visualizer.
FlaskReactPostgreSQLRedisKafka
Data plane
Mesh
A Python sidecar process injected next to every agent pod, mediating all inter-agent traffic over mTLS using the A2A JSON-RPC protocol. Interceptor pipeline runs auth, policy, compliance, PII redaction, guardrails, audit, rate limiting, and resilience.
PythonmTLSA2AKubernetesHelm
Capabilities
AI enforcement on the network layer
Everything you'd build over six quarters of platform work — without the six quarters.
Sidecar mesh, mTLS by default
Sidecar injected next to every agent pod. All inter-agent traffic over mTLS using the A2A protocol with auto-rotating certs.
Real-time guardrails
Pre/post-processing interceptors with regex, vector DB, LLM-as-judge, and ML classifiers. Push policy from the registry, evaluated inline.
Identity & policy
mTLS cert CN, JWT, API key auth. Priority-ordered allow/deny with wildcard matching. Governance status enforced on every call.
Compliance built in
Sovereignty zones, data classification, GDPR consent enforcement. PII detection with Presidio or regex — redact, block, or warn modes.
Observability & audit
OpenTelemetry instrumentation, W3C trace context, hash-chained tamper-evident audit log, live Socket.IO mesh visualizer.
Resilience & traffic
Circuit breakers, retries with backoff, rate limiting, fault injection, weighted routing, failover. Production-grade traffic management.
Multi-cluster HA
Active-active across clusters with PostgreSQL replication and event bridge. Multi-registry failover with health-based promotion.
Kubernetes native
Mutating admission webhook injects sidecars automatically. Helm chart with values overlays. Calico-backed NetworkPolicy enforcement.
Bring your own agents
Works with OpenClaw, LangChain, LangGraph, CrewAI, plain Python, or anything that speaks HTTP. Recursant adds identity and policy — your code stays untouched.
Quick start
From zero to a running mesh in one command
Recursant runs in Kubernetes — Kind for local dev, any cluster in production.
# 1. Configure secrets (LLM API key)
cp .env.sample .env
$EDITOR .env
# 2. Bring everything up — Kind cluster, build, deploy, smoke test
./scripts/install.sh
# 3. Open the registry UI
open http://localhost:8030Recursant is released under MIT license. You can use it commercially, fork it, or change it as you like. Contributions are more than welcome.
Every agent will need a supervisor
Star the repo, file an issue, send a PR. Recursant is better with more eyes on it.
ajensenwaud/recursant