The AgentRun console displays real‑time analytics for autonomous agents.
AGENTRUN OPEN-SOURCES SELF-IMPROVING AGENT PLATFORM
*GREP.AI, a YC F26 startup, releases AgentRun as a full managed‑agents platform. The move promises self‑improving AI agents, extensive toolkits, and a hosted or self‑hosted option for developers.*
GREP.AI, a Y Combinator Fall 2026 cohort, has pushed AgentRun from internal harness to an open‑source, managed‑agents platform. The announcement landed on Hacker News with a link to a detailed blog post. AgentRun promises self‑improving agents that consume fewer tokens while delivering richer outcomes. The codebase now includes web‑grounded workflows, sandboxed environments, and a library of thousands of tools. Detailed analytics let operators audit every decision. The move signals a shift toward commoditized AI agent infrastructure.
What does AgentRun offer out of the box?
AgentRun ships with self‑improving agents that refine their own code. It supports web‑grounded workflows that pull live data into reasoning loops. The platform bundles sandboxes and computer environments for safe execution. Thousands of tools and integrations are pre‑wired, letting agents call APIs, scrape sites, or manipulate files without custom code. Detailed analytics track token usage, decision paths, and performance metrics, giving operators visibility into agent behavior.
How can developers integrate their own models?
AgentRun accepts any harness that implements the ACP protocol. Users can drop in Pi, Claude Agents, Codex, OpenCode, or bespoke models. The architecture isolates each agent in its sandbox, preserving security while allowing cross‑communication. Developers simply point the platform at their model endpoint and define the toolset. No code rewrite is required; the platform translates ACP calls into actionable tasks. This flexibility lowers the barrier for teams to experiment with emerging LLMs.
What deployment options are available?
The team provides a one‑click install on Render for self‑hosting. A valid OpenRouter key unlocks the full stack on private infrastructure. For rapid trials, a hosted console lives at console.agentrun.ai. Both paths deliver the same analytics suite and sandbox isolation. Self‑hosting gives firms control over data residency; the hosted service offers instant scalability. The dual model lets organizations choose speed or sovereignty without sacrificing features.
What feedback is the team seeking?
GREP.AI asks users to critique the architecture outlined in the README. They want to know which integrations are missing and how the sandbox performance holds under load. The open‑source release invites community patches to improve stability and expand tool coverage. Direct feedback will shape the next iteration of the managed platform. The call for input underscores the project's collaborative ethos.
Frequently asked questions
What is AgentRun?
AgentRun is an open-source, managed agents platform built by GREP.AI that includes self-improving agents, web‑grounded workflows, sandboxes, thousands of tools, and detailed analytics.
How can developers deploy AgentRun?
Developers can self‑host AgentRun with a one‑click install on Render using an OpenRouter key, or they can use the hosted version at console.agentrun.ai.
Which AI models are compatible with AgentRun?
AgentRun works with Pi, Claude Agents, Codex, OpenCode, and any other harness that supports the ACP protocol.
AgentRun’s open release could accelerate the race to deploy autonomous AI agents at scale. By exposing a turnkey platform, GREP.AI invites both startups and incumbents to test, extend, and monetize self‑improving agents. The dual deployment model—hosted and self‑hosted—covers the spectrum from rapid prototyping to regulated production. As the community builds on the code, the platform may become the de‑facto backbone for next‑generation AI operations.
Sources: Hacker News post – Show HN: AgentRun – open-source, Self-improving Managed Agents