Rebuno
Rebuno is an open-source execution runtime for running AI agents in production. It makes tool and LLM calls durable, records execution history, and applies policies that can allow, deny, or pause calls for human approval.
At a Glance
- Teams operating AI agents in production
- Developers and platform engineers building agent infrastructure
- Organizations needing auditability, policy controls, and human approvals for AI actions
- Users of LangChain, CrewAI, Pydantic AI, Vercel AI SDK, and Mastra
AI Tools by Rebuno
(1)
Rebuno
AI Agent Execution Runtime
Discussions
No discussions yet
Be the first to start a discussion about Rebuno
Latest News
Show HN: Rebuno - An open-source runtime for production agents
Rebuno: an execution runtime for production agents
Rebuno core release v0.26.0
Rebuno core release v0.25.0; includes returning 404 for events of nonexistent executions
Products & Services
The open-source runtime and kernel for production agents. It dispatches work to agent HTTP services, records tool and model calls as steps, stores execution state, replays completed steps after interruption, and enforces per-agent policies.
Python package for defining agents and tools, connecting model calls to Rebuno, recording local steps, inspecting executions, and resolving approvals.
TypeScript SDK for defining and serving agents, declaring tools, making durable LLM calls, recording steps, and interacting with executions and approvals.
A TypeScript web dashboard for viewing executions, steps, events, and agent activity.
Market Position
Rebuno positions itself as an execution runtime rather than a workflow engine or another prompt-level guardrail. Its differentiators are durable per-step replay, explicit handling of unknown tool outcomes through idempotency, and policy enforcement outside the model prompt; the introductory article contrasts it with workflow engines, framework checkpointing, and guardrails embedded in application code.
Leadership
Founders
Nathan Wong
Identified by PyPI as Rebuno's author and by GitHub as the nathanvvong account associated with the repository; the Rebuno Show HN announcement was posted by nnww, who said he had been building Rebuno. Public sources reviewed do not provide earlier roles or companies.
Executive Team
Nathan Wong
Project author and maintainer
Listed as the author of the Rebuno Python package and associated on GitHub with the Rebuno repository; public sources reviewed do not give a fuller executive biography.
Founding Story
Rebuno was started to address the operational problems of running AI agents in production: interruptions, repeated side effects, and guardrails that depend only on prompts or application code. Its initial vision is a framework-independent runtime where each model or tool call is recorded as a durable step, replayed safely after interruption, and evaluated by policy before execution.
Business Model
Revenue Model
The core project is free to self-host and released under the MIT license. Rebuno's website says a hosted version is being worked on and invites users to join a waitlist; no public hosted pricing or revenue figures were found.
Target Markets
- Teams operating AI agents in production
- Developers and platform engineers building agent infrastructure
- Organizations needing auditability, policy controls, and human approvals for AI actions
- Users of LangChain, CrewAI, Pydantic AI, Vercel AI SDK, and Mastra
- Operating AI agents in production
- Durable multi-step agent workflows that must recover from crashes or restarts
- Agents that perform side effects such as sending email or other tool actions
- Human-in-the-loop approval for sensitive agent actions
- Centralized policy enforcement and audit trails across multiple agent frameworks
- Self-hosted execution infrastructure for teams building AI agents