OPEN SOURCE FROM JUVANT

Open source — for thehumans and the agents.

The OSS arm of Juvant. Frameworks, MCP servers, and tooling we use ourselves — released under MIT so you can run your own AI-powered agent stack.

Star juvant-os Browse repos →

Featured repos

License default

MIT

Docs license

CC0

Updated

Continuously

01 — WHAT WE PUBLISH

One framework, a few servers, the tools we built around them.

●FLAGSHIP · juvantlabs/juvant-os

The skill-first multi-agent OS for Claude Code.

Open `claude` in a folder, say "Initialize Juvant OS for <your-company>", the Skill orchestrates the company. No CLI, no daemon, no installation.

frameworkclaude-codemulti-agentMIT

View on GitHub

MCP SERVERS

juvantlabs/m365-graph-mcp-server

Microsoft Graph integration for calendar, mail, and tenant ops.

MCP server for Microsoft Graph API — manage calendar events, email, teams, and organizational structure.

mcpmicrosoft-graphintegration

License: MIT

TOOLS & LIBRARIES

juvantlabs/engram

LLM-cached browser automation with Playwright.

Python-native browser automation with cached LLM-resolved selectors. One LLM call per UI version, not per run.

toolautomationpythonplaywrightllm

License: MIT

juvantlabs/juvant-tools

CLI toolbox for Juvant operations.

Operational utilities and deployment helpers used by the Juvant team.

toolclioperations

License: MIT

02 — HOW WE WORK

Six principles. Visible in every commit.

These aren't aspirations — they're the operating manual we hand to every contributor, human or otherwise. If a PR reads like one of these, it's likely to land.

WORK IN THE OPEN

Public ADRs. Public roadmap.

Architectural decisions live in handbook/docs/adr — visible, dated, reversible. The framework roadmap is a public board. If you want to know why we did something, the receipt is in the repo.

EAT YOUR OWN DOGFOOD

juvantio runs on juvant-os.

The same framework that ships under MIT here is what runs the company. When we change the abstractions, we feel it first. No shelf-ware, no demo-only paths.

PERMISSIVE BY DEFAULT

MIT for code. CC0 for docs.

Use it, fork it, ship it. Attribution welcome but never required for the documentation. We want the ideas to spread further than the brand does.

AI-ASSISTED, ON THE RECORD

PRs from agents are welcome.

We use Claude to write a lot of this code. So can you. The conventions are in CONTRIBUTING.md — disclose the assistance, sign the commits, keep the diffs small enough to read.

COORDINATED DISCLOSURE

Security, both ways.

Report vulnerabilities to security@juvant.io — we follow the process in handbook/docs/security/. When we audit a community alternative and find issues, we publish the report as a public gist after coordinated disclosure.

CALM, NOT QUIET

No hype. Plain English.

We won't tell you this revolutionises anything. We'll tell you what it does, what it doesn't, and where the rough edges are. A README that respects your time is the smallest form of respect we can offer.

03 — WRITING & ADRs

The receipts.

ADRMay 3, 2026

ADR 0001 — GitHub account and organization structure (juvantlabs / juvantio)

Codifies structural decisions implicit in the two-namespace model (juvantlabs for OSS, juvantio for company operations), operationalized through features and dogfood actions.

Read →

ADRMay 3, 2026

ADR 0002 — MCP abstract roles + binding policy

Codifies patterns in agent tool matrix and MCP role binding, operationalized via FEAT-011 (Finom) and FEAT-012 (Aruba) re-scoping decisions.

Read →

ADRMay 4, 2026

ADR 0003 — Scope boundaries for MCP servers

Defines threat-model boundaries for MCP servers. Each server wraps single capability scope with clear delineation; emerged from m365-graph-mcp-server decisions.

Read →

ADRMay 4, 2026