Runta

Runta

Runta is an execution layer for AI agents. Instead of bolting security onto a generic sandbox, Runta rebuilds the systems-software layer so an autonomous agent runs inside a controlled operating-system environment, locally or in the cloud, with a stateful computer of its own. Runta governs an agent's access to the OS, network, file system, and credentials, enforcing policy in real time so agents only touch approved systems and data. It targets four production problems — wasted tokens, idle compute, exposed raw credentials, and open network egress — and keeps a full audit log of every action. The framing from its investors is that agents just want a computer, and Runta provides a secured, stateful one.

Production credibility: Runta is well-documented rather than vendor-only: multiple outlets report a $20M seed round led by Andreessen Horowitz at a valuation above $100M in mid-2026, and a16z partner Martin Casado publicly backed it. Founder Guanlan Dai previously led edge-proxy technology at Cloudflare and core-proxy engineering at Kong, which is directly relevant pedigree for owning the agent-infrastructure layer. Runta publishes an SDK and developer documentation, and maintains X, GitHub, LinkedIn, and Discord channels. It is early — a seed-stage company only recently public — so hosted-product maturity and availability may still be limited.

Key Features

• An isolated, stateful operating-system environment per agent
• Real-time policy enforcement over OS, network, file-system, and credential access
• Per-agent spending caps to prevent runaway token and compute costs
• A full audit log of every action an agent takes
• Credential brokering so agents never handle raw secrets
• Network egress controls with allow-listed access
• Local or cloud execution with a developer SDK
• Built for production agents that execute code, touch credentials, or make external calls

Ideal Use Case

Runta fits engineering and platform teams at companies deploying autonomous AI agents into production that need isolation, governance, and cost control. It is aimed at anyone whose agents execute code, touch credentials, or make external calls and who needs auditability and hard guardrails around that. It is less relevant to a hobby project or a single-shot agent with no sensitive access, where a lightweight sandbox is enough and Runta's governance layer is more than the task requires.

How Runta differentiates

Runta positions itself as core systems software rebuilt for agents rather than another code-execution sandbox cloud, so the emphasis is a purpose-built stateful execution layer with credential brokering and egress control built in rather than added on. Its founder's Cloudflare edge-proxy and Kong core-proxy background is the credibility behind owning that layer, and its a16z-led seed marks it as a well-backed entrant in agent infrastructure.

FAQ

Q: What is Runta? A: Runta is an execution layer that gives each AI agent an isolated, stateful computer with built-in sandboxing, access controls over OS, network, and credentials, spending caps, and full audit logging.

Q: How is it different from a code sandbox? A: Rather than adding security to a generic sandbox, Runta rebuilds the execution layer so credential brokering, egress control, and policy enforcement are native, aimed at production agents rather than one-off code runs.

Q: Who funds Runta? A: Multiple outlets report a $20M seed led by Andreessen Horowitz at a valuation above $100M in 2026, with a16z partner Martin Casado publicly backing it.

Q: What background does the founder have? A: Founder Guanlan Dai previously led edge-proxy technology at Cloudflare and core-proxy engineering at Kong — directly relevant infrastructure pedigree.

Q: Is there public pricing? A: No self-serve pricing is published; Runta is enterprise infrastructure arranged through sales, though a developer SDK and documentation are available.

tl;dr

Runta is an execution layer that gives AI agents an isolated, stateful computer with real-time controls over OS, network, and credential access, spending caps, and full audit logging. It rebuilds systems software for agents rather than securing a generic sandbox, and is backed by a $20M a16z-led seed with a founder from Cloudflare and Kong. Best for teams putting autonomous agents into production; it is early and enterprise-sold.