aos

What AOS is for

AOS stands for Agent Operating System, and the README describes it as a Web first, multi tenant workspace for building agent workflows around a single recoverable session. The design brings conversation, evidence based research, data questions, memory, files, Skills, MCP, task recovery, and external bot channels into one product surface. The stated purpose is to make agent work inspectable and recoverable rather than opaque and disposable. The README lists concrete capabilities: one recoverable agent session for normal questions, tool use, long running work, and follow up context; deep research that gathers and checks evidence with bounded execution and a usable fallback answer; data work using datasource aware NL2SQL, semantic retrieval, SQL knowledge, and attribution workflows; mobile agent access through Bot Gateway adapters for supported chat platforms and notification delivery; and progressive engineering design that turns a product request into a reviewable core design, then an editable implementation plan and task breakdown for an external coding agent. Skills and MCP are described as optional extensions that can be installed, inspected, and governed. The project status note is candid: AOS is an active source release under regression testing, and the repository contains the source tree, documentation, tests, and packaging scripts. Prebuilt offline archives are kept out of Git history and attached to GitHub Releases only when a build is ready. The README warns that some integrations need provider credentials and external services.

Architecture and data model

The README presents an architecture diagram in text. A Web UI or external bot connects to a Rust web server, which owns the SQLite platform state. From there the server reaches the agent runtime and recoverable sessions, the model providers and optional web search extensions, and the datasources, Skills, MCP, and repository adapters. The platform database is SQLite and belongs to the AOS control plane, while business databases are configured as external NL2SQL datasources and are not required to boot AOS. A single AOS process owns one local data directory. The repository map divides the code into a Rust workspace for the API, runtime, agent orchestration, and web server; a webui directory for the React and Vite application; a docs directory for architecture, deployment, security, integration, and test documentation; and eval and examples directories for deterministic fixtures and small integration examples. The configuration boundaries section explains that model credentials are stored per tenant through the API Keys surface and are never committed, that search providers and MCP servers are optional, and that Skills are treated as untrusted input during installation and subject to safety checks and permission boundaries. Long running tasks persist state and events so a client reconnect can recover context. The README is explicit that the menu map is a product map, not a promise that every capability is enabled in every deployment.

Running and testing the source

The shortest supported path for contributors is Docker. The README shows a generate env script followed by docker compose up, then opening http://localhost:3000 to finish setup, and it tells the user to add an enabled chat scoped API credential in System then API Keys before asking the model to do provider backed work. Native development uses a setup environment script and an aos start script, with deployment guides linked for running beyond localhost. Development checks run cargo fmt, cargo check for the web server, and cargo test across the workspace with all features, then npm ci, npm run typecheck, and npm run build ci in the webui directory. Targeted test guides cover the Bot Gateway, data attribution, NL2SQL configuration, and the progressive engineering design workflow. Release artifacts are described as a separately built AOS Offline distribution for macOS, Linux, and Windows x64; the archives are not committed, and maintainers build, validate, and publish them under GitHub Releases with a checksum and upgrade notes. Until that happens, the source workflow is the supported route. The README also links a security policy and contributing guide. With 35 stars in the metadata, AOS is an early stage project, but its documentation covers architecture, capability profiles, watchdog design, bot contracts, and NL2SQL design in dedicated files.

Editorial conclusion

AOS is licensed under the Apache 2.0 license and its most recent commit was on 2026-08-26.