Platform

Runtime governance built for production AI

Every agent run gets its own sandbox: network egress allowlisted, CPU and memory capped, file I/O scoped. Violations are blocked and logged before they reach your infrastructure.

Read the docs See pricing
Runta Runtime
Policy Engine
Audit Logger
Agent Sandbox
Quota Enforcer
Egress Checker
Trace Emitter
Your Infrastructure
API Gateway
Data Stores
External APIs
Four pillars

What Runta governs

Runta wraps every agent invocation in four overlapping controls. Each one is independently auditable and tunable per agent role or workspace.

Sandbox isolation

Each agent run executes inside a lightweight container with no shared state. File system access is scoped to declared paths; network is fully blocked by default.

Resource quotas

Define per-agent CPU seconds, memory caps, token budgets, and wall-clock timeouts. Quota violations are enforced at the kernel level, not the application layer.

Egress allowlisting

Declare the exact hostnames and IP ranges an agent may contact. Any outbound connection not on the allowlist is blocked and recorded before the TCP handshake completes.

Immutable audit logs

Every tool call, egress attempt, quota event, and output token is appended to a tamper-evident log. Entries cannot be modified or deleted by agent code or operator keys.

SDK integration

Drop into any stack in under an hour

Runta exposes a thin Python SDK. Wrap your agent with a single context manager: your existing LangChain, AutoGen, or custom orchestration code runs unchanged inside the sandbox.

  • Python 3.9+ SDK, zero compile step
  • Works with any agent framework
  • Policy-as-code via YAML config
  • OpenTelemetry trace export
SDK usage
import runta
from my_agent import ResearchAgent
 
# Load policy from repo
policy = runta.Policy.from_yaml("runta.yaml")
 
with runta.sandbox(policy=policy) as sb:
  agent = ResearchAgent(sandbox=sb)
  result = agent.run("Summarise Q2 risks")
 
# Audit log auto-committed, sandbox torn down
audit = sb.audit_log()
print(audit.egress_blocked, audit.quota_events)
Observability

Trace every decision at every layer

The Runta dashboard streams live trace events: tool calls, quota ticks, egress checks, and audit-log commits. Filter by agent run, workspace, or policy rule.

Live trace stream

Sub-millisecond event latency from sandbox to dashboard.

Policy diff viewer

See exactly which rule fired on each blocked event.

Log export

Export tamper-evident logs to S3, GCS, or any SIEM in minutes.

Alert routing

PagerDuty and Slack webhooks on quota breach or egress block.

Ready to govern your agents?

Get the sandbox up in your environment today. The Sandbox tier is free, no card required.

Start with quickstart Compare plans