How are policies versioned?

Policies live in Mesh configuration with audit on change; test in staging namespaces before fleet rollout.

What happens on violation?

Configurable: block the action, alert only, or halt the agent via kill switch integration.

Can policies reference external IAM?

Enterprise deployments integrate identity and data classification sources; scope depends on your Mesh configuration.

AXME MESH

Define what your agents are allowed to do — and enforce it automatically

AI agents need guardrails that go beyond code review: runtime limits on what tools they can call, what data they can access, how much they can spend per hour.

AXME Mesh lets you write policy rules (rate limits, tool allow/deny lists, budget caps, data scope) that are enforced at runtime — not bolted on after the fact.

Start free Read docs

Your agent called an API it was not supposed to touch. The restriction was in the system prompt — prompt injection bypassed it. Gateway-enforced policies block the call before it reaches the tool, not after damage is done.

Guardrails at runtime, not only in prompts

Telling an agent "do not exceed $50/hour" in the system prompt is not enforcement. Code review cannot catch every tool call in production. Policies must evaluate before damage — rate limits, tool allow lists, data scopes, and budgets.

AXME Mesh evaluates YAML policies at runtime with inheritance from fleet defaults to per-agent overrides. Violations alert and can trigger kill switch or intent rejection.

CAPABILITIES

How it works.

Rate limits

Calls per minute per agent.

Budget caps

Spend limits with hard stops.

Tool allow/deny

Which APIs agents can invoke.

Data scope

Which records and tenants.

DEEP DIVE

Production patterns.

Runtime enforcement

Block before damage — not after.

Violation alerts

Notify on breach.

Fleet inheritance

Defaults + per-agent overrides.

Prompt rule vs gateway policy

Prompt only

# system: never call send_email
# agent calls send_email anyway

Mesh policy

policies:
  tools: [crm.read]  # send_email denied at gateway

Policy example (YAML)

policies:
  - agent: support-bot
    tools: [email.send, crm.read]
    budget_usd: 50/day
    rate: 100/min

Policy primitives

Rate limits cap calls per minute per agent or namespace. Budget policies tie to cost-control hard stops. Tool allow/deny lists restrict which APIs an agent may invoke. Data scope rules limit which records or tenants a retrieval step may touch.

Start from templates for common patterns — support bots without send-email, research agents without production DB write — then customize per team.

Common questions

How are policies versioned?: Policies live in Mesh configuration with audit on change; test in staging namespaces before fleet rollout.
What happens on violation?: Configurable: block the action, alert only, or halt the agent via kill switch integration.
Can policies reference external IAM?: Enterprise deployments integrate identity and data classification sources; scope depends on your Mesh configuration.

Ship your first durable agent — in under 10 minutes.

Free tier. No credit card. Self-host or hosted — your choice.

Start free now Read the docs