What Runta is
Runta is an execution-infrastructure platform for AI agents, which it calls “the execution layer for AI agents”: durable, isolated runtimes for agents, with controls for token spend, credentials, network access, and compute handled at that layer. The company was founded by Guanlan Dai. See the Runta website.
Capabilities and onboarding
Per the website, Runta’s capabilities fall into three groups. Spend: prompt compression to trim redundant context, and Token X-ray for inspecting captured tool inputs and outputs to spot repeated calls. Control: protected credentials replace raw secrets with usable credential stubs, with real values injected only at the authorized network boundary; egress is governed by allow or deny rules on exact and wildcard hosts. Scale: idle runtimes are suspended automatically and resumed on demand while preserving state, and memory auto-scales within a configured cap.
Onboarding is to install the Runta CLI, authenticate, create a runtime, and run the agent inside it — the controls apply without modifying the agent itself. The site also links documentation, pricing, an OpenAPI spec, and a blog, with a free trial via dashboard.runta.com.
Funding and FrontierHarness Eval
The official about page lists investors including World Labs (Martin Casado) and Jeff Dean, Fei-Fei Li, Ali Ghodsi, Ram Shriram, and Thomas Wolf. According to reporting by The Next Web and others, Runta raised a $20 million seed round led by a16z in July 2026 at a valuation above $100 million (per The Information). On September 1, 2026, Runta published FrontierHarness Eval (FrontierHarness v1.0) on its blog: 9 coding agent harnesses evaluated on 30 tasks across 12 configurations — 360 runs in total — under a unified model (Kimi K3) and runtime, scored by a deterministic verifier. See the official blog post. More background is on Guanlan Dai’s page.
Discussion in the show
In Weekly #001’s chapter “Harness 评测：完成率、Token 成本与”斩杀线”” (“Harness evaluation: completion rates, token costs, and the ‘culling line’”), Yang Pan relayed the day’s news: “Guanlan’s team, that is Runta, just released an evaluation called Harness Eval,” comparing harnesses by task completion rate, token cost, and other angles. The host’s reading of the results is a relay that broadly matches the official blog data; it is not a statement from Runta. See the episode 001 chapter in the Chinese transcript; an English transcript is not available.
Frequently asked questions
What is Runta?
An execution-infrastructure platform for AI agents: durable, isolated runtimes with controls for token spend, credential protection, network governance, and resource scaling. See runta.com.
Who founded Runta?
Guanlan Dai, previously an early engineer on Cloudflare’s Edge team and a founding engineering lead at Kong, per the official about page.
How much does Runta cost?
The site offers a free trial via dashboard.runta.com, and plans are described on the official pricing page.
What is FrontierHarness Eval?
A benchmark Runta published on September 1, 2026, comparing 9 coding agent harnesses on 30 tasks under a unified model, reporting completion rates and token costs — see the official blog post.