AgentOS.
Rent the intelligence. Own the loop.
AgentOS owns the loop around the coding agents you already run. Describe intent in a few sentences; AgentOS plans it into tasks, drives Claude Code or Codex as interchangeable workers, verifies every result deterministically, and keeps the plan, memory, trust record, and handoffs in your repo.
cargo install agentos-build01 --lockedA Loop That Forgets Is Just a Wrapper.
Most developers and teams run AI agents through ephemeral terminal chats or closed vendor wrappers. When a session terminates, the reasoning is lost, task progress cannot be verified deterministically, and code changes lack an inspectable audit trail.
No Vendor Lock-In
Swap frontier LLMs and worker runners without changing your task pipelines or plan structures. Treat intelligence as interchangeable commodity compute.
Deterministic Gates
Automated evaluators test against contracts, test suites, and git diff invariants before acceptance. Completion is verified, never self-reported.
Repo-Local Knowledge
Durable facts, architectural decisions, and run checkpoints stay in plain Markdown under .agents/memory/ in your versioned repository.
Engineered for Autonomous Execution.
Six architectural pillars designed to make AI coding agents dependable, inspectable, and repo-native.
Loop Engineering & Task Planning
Translates intent into immutable draft proposals and inspectable semantic diffs before applying any changes to your repository.
Interchangeable Worker Dispatch
Drives Claude Code, Codex, and other LLMs as swappable commodity workers. You retain full control over the execution harness.
Deterministic Verification
Evaluates generated code against test suites, schema contracts, and scope drift from git diffs so completion is never self-reported.
Repo-Local Knowledge Mesh
Durable facts, architectural decisions, and run checkpoints stay in .agents/memory/ and .agents/runs/ without prompt bloat.
Guarded Feedback Cycles
When a validation fails, AgentOS captures the failure log and feeds it into the next attempt, bounded by configured loop limits.
Single-Binary Rust CLI
Zero heavyweight dependencies. Install via cargo install agentos-build01 --locked and run seamlessly on Linux, macOS, and Windows.
How AgentOS Runs.
From human intent to verified code commits — a deterministic loop engineered for repeatable, safe autonomous workflows.
Describe Intent
Describe what needs to be built in natural language. AgentOS opens a planning channel and generates immutable draft revisions with inspectable semantic diffs.
Worker Execution
Drives Claude Code, Codex, or local models as interchangeable workers. Workers execute tasks without prematurely polluting active repo state.
Deterministic Verification
Evaluators run tests, contracts, and git diff invariants. Failures are captured and injected into bounded feedback cycles before acceptance.
Repo-Local Checkpoint
Approved tasks commit directly to your git history, updating durable memory in .agents/memory/ and preserving complete provenance.
AgentOS Frequently Asked Questions.
Everything you need to know about adopting, running, and orchestrating AgentOS in production environments.
AgentOS is Build01's flagship open-source autonomous runtime. It owns the loop around the coding agents you run — planning intent into tasks, driving interchangeable workers (like Claude Code or Codex), verifying results deterministically, and storing all memory in your repository.
Ready to Own Your Agent Loop?
AgentOS is open source, built in Rust, and available on crates.io today. Install the CLI, inspect the codebase, or connect with our team to discuss runtime integration.