Find your next open-source project

Search by name, purpose, or keyword — or paste a GitHub repository URL

vstorm-co/agenticos

AI Engineering

Self-hosted platform for teams to build, publish and govern AI agents with shared chat, RAG, tools and budgets.

351
60
Python
Apache-2.0
3 months old

What it does

AgenticOS is an Apache-2.0, self-hosted layer for building, sharing and governing AI agents inside a company. Teams write an agent's instructions in the browser, choose a model, enable tools and publish the agent to web chat, Slack, Teams, Discord, Mattermost, Telegram, a website widget, a hosted page, the HTTP API or a WebSocket. Administrators set roles and department groups, per-agent and per-organization budgets, approval policies for sensitive tools, and a hash-chained audit log records every run.

The design separates code from configuration: engineers register capabilities in typed Python, while non-developers compose agents from what is registered. It is built on FastAPI, Pydantic AI, PostgreSQL with pgvector, Redis and Next.js architecture section. Retrieval works by parsing uploaded or synced documents, chunking and embedding them into pgvector, then answering with citations; local parsers and Ollama embeddings are supported for a fully local setup.

Quick start

The fastest developer path is Docker Compose plus a model provider. On macOS or Linux, run:

curl -fsSL https://raw.githubusercontent.com/vstorm-co/agenticos/main/scripts/quickstart.sh | bash

On Windows, run the same command in WSL2 with Docker Desktop's WSL2 integration enabled. The installer asks for a model provider and key, a login and an organization name, pulls the published images and starts a deployment. The documented minimum is 4 vCPU and 8 GB RAM; then open http://localhost:3000 and sign in with the chosen login local quickstart. The installer can be inspected first, and --check verifies prerequisites without installing quickstart script.

After signing in, the first walkthrough uploads a handbook, asks questions and checks answers against cited sources document-agent tutorial.

Alternative entry points: a desktop app connects to an existing workspace without local Docker desktop guide; IT can deploy a company server from a pinned release and bootstrap the first organization deployment guide; one-click templates exist for RepoCloud and Railway, though the Railway template pins an older release and omits the local code sandbox.

Requirements and limits

  • Hosting: the supported production deployment is Docker Compose on a single Linux host with 4 vCPU, 8 GB RAM, Docker Engine 24+, Compose 2.24+ and make; two API workers are suggested for a team of ten.
  • Operations: self-hosting needs someone to own updates, backups, access and external services. Backups must include the database and the vault key in backend/.env.
  • Models: 28 providers are listed, including hosted APIs, Azure/Bedrock/Vertex through your cloud contract, the ChatGPT subscription via Codex, and local Ollama, LiteLLM, vLLM or LM Studio.
  • Identity: email/password with magic links, Google, OIDC (Entra ID, Okta, Keycloak and others), LDAP and Kerberos. MFA comes from your identity provider; there is no native MFA, SAML or SCIM yet permissions.
  • Data boundaries: source-system permissions are not mirrored automatically, so an agent can expose bound knowledge to its users. Self-hosting the console does not make every model, parser or tool local — each outbound destination must be reviewed security.
  • Mobile: the browser console works on phones, but native iOS/Android apps are development builds with no app-store release.

HotRepo's take

The repository was created at the end of July 2026 and on 2026-11-16 recorded 345 stars and 58 forks, gaining 56 stars in that day's Python daily list, where it reached 12th. That is a single day of data, so it shows a burst of attention rather than a sustained trend, and the project has no recorded best rank on the all-language daily board.

The scope is broader than most agent frameworks: chat channels, MCP catalog, budget checks, approvals and an audit log are all part of the same server, and the README is explicit about what is not yet covered, such as permission mirroring and native MFA. For teams already running Python and PostgreSQL infrastructure, the Docker Compose path and the documented capability extension points make the trade-offs easy to evaluate.

Written by AI from the project's README, updated . The project's own docs are the source of truth.

Hacker News

One place to build, run and govern your company's AI agents
▲ 1 · 0 comments · September 7, 2026

You might also like

  1. agentes-chat/agentes-chatSelf-hosted team chat where people and AI agents share channels, DMs and approvals.TypeScript183
  2. paperclipai/paperclipRuns a team of heterogeneous coding agents like a company: goals, org chart, budgets and audit trails in one dashboard.TypeScript106,775
  3. ahmadrosid/nakamaSelf-hostable multi-tenant platform for teams to run, share and govern AI agents with tools, memory and messaging channels.TypeScript773
  4. Shubhamsaboo/awesome-llm-appsA collection of runnable AI agent, RAG and agent-skill examples for developers learning to build LLM applications.Python142,990
  5. mcpjungle/MCPJungleSelf-hosted gateway that registers MCP servers once and exposes their tools through a single endpoint to AI clients and agents.Go1,575