Back to directory
AI & ML · AI infrastructure / MLOps / evals

Runta

Runta is your execution layer for AI agents.

Building @Runta, the execution layer that controls what AI agents can actually do. Prev @Cloudflare and @Kong.
San Francisco, CA4.7K followers
TLVC Rating
Hook
Editing / Creativity
Copy
Sentiment of launch
Distribution strategy
Community Rating
★★★★★
No ratings yet
Your rating
Sign in to rate this launch.

About

Runta is an execution layer that sits underneath AI agents to control what they can actually do while they run , aimed at the developers and platform teams now handing agents credentials, budgets, and access to production systems. The launch pairs the general availability push with a $20 million seed round led by a16z , giving the company capital to scale a runtime that treats agent behavior as something to be bounded at execution time rather than at the code review stage. The product focuses on four failure modes the team sees in real agent deployments: token waste, idle compute, raw credentials, and open egress . Runta compresses oversized tool outputs before they hit the model, auto-suspends and resumes runtimes when agents go idle, injects secrets so agents never touch raw credentials, and governs network egress with a policy layer. Every action the agent takes is captured at the OS level, so operators can trace what happened when something breaks. A TypeScript, Python, and CLI SDK lets teams wrap an existing agent (Codex, for example) in a Runta runtime with a few lines. Runta is founded by Guanlan Dai, previously at Cloudflare and Kong , where he was one of the earliest engineers on Cloudflare's Edge team and later led the Edge Platform team working on CDN, WAF, and the front-line request path, helping scale that edge to more than 10 trillion requests per month serving over 2.5 billion users . At Kong he was a founding engineering leader who built and led the teams behind Kong Gateway, Kong Cloud, the Kubernetes Ingress Controller, and Kong's AI infrastructure products as the company grew to 700+ enterprise customers running over 400 billion API calls per day . The broader team has contributed to Tokio and OpenResty , and the seed round was led by Martin Casado of Andreessen Horowitz, with participation from Jeff Dean, Fei-Fei Li, Ali Ghodsi, Ram Shriram, and Thomas Wolf .
Tags
<500KAI agentSeedExplainerB2BGlobalFunding announcementFounder-led
Comments (11)
Sign in to join the discussion.
Priya Khatri4d ago

The framing of runtime constraints vs write-time constraints is genuinely the cleanest one-liner I've read this quarter. Someone made you rewrite that tweet six times and it shows.

Mikael Sund4d ago

The landing page hero has the button and the headline fighting for the same optical weight. Fix the leading on the subhead and it sings.

Tomasz Wróbel4d ago

Curious how you handle streaming tool calls when the policy engine wants to block mid-execution. Also, do you expose webhooks for policy violations or just logs?

Kenji Otabe4d ago

Every agent infra startup this year is pitching 'the execution layer' and I still cannot tell which of you actually runs the sandbox vs just proxies OpenAI calls.

Borys Dovzhenko4d ago

Would be neat if policies could settle on-chain for audit trails across orgs. Not saying token, just saying receipts.

Adaeze V.4d ago

Been circling this exact wedge since spring, told three of my LPs Guanlan was the one to back. Feels good to be right in public for once.

vikram4d ago

Retention question nobody wants to ask: what stops a customer from ripping this out once they've written their own policy DSL in a weekend? Interesting either way.

Lola Adeyemi4d ago

The launch video's pacing is aggressive in a good way, no 12 second intro of abstract shapes. Whoever cut it earned their invoice.

Chen Xi4d ago

Cool tech, but the real question is your deployment story. Are teams running this as a sidecar, a gateway, or an SDK, because each one has completely different sales motions.

Hana E.4d ago

Intern question: if the agent gets blocked, does it know why, or does it just retry the same action in a loop until it burns its context window?

Moremi O.4d ago

Would love to hop on a quick call before Wednesday if you're doing press this week. Just a few questions about the a16z process and what's actually shipping vs vaporware.