AGPLv3 Community Edition Commercial License Available Security Policy Governance Roadmap Citation Defensive Publication

Responsible AI
for everyone.

Nunneri is a runtime-neutral agentic AI control plane: canonical assets compile into provider adapters, runtime contracts, human-blocking gates, durable state, and Graph Studio execution.

Start with Community Edition ↓ ★ View on GitHub →

Originated by Suranku and Yamini. Stewarded by the Nunneri Core Team.

Runtime-neutral architecture

One contract. Many providers. Many runtimes.

Nunneri keeps prompts, workflows, approvals, tenant policy, and run evidence in one neutral control plane, then projects that contract into assistants, runtimes, and Graph Studio.

Canonical assets Agents, skills, commands, workflows, and repo context under assets/.

→

Nunneri Runtime Contract Generated neutral JSON for phases, edges, approvals, context, and observability hints.

→

Execution surfaces Provider files, runtime exports, API graph definitions, and Graph Studio metadata.

Providers

Claude CodeCLAUDE.md and .claude/ assets.

CodexAGENTS.md and .codex/ assets.

Gemini + open sourceGEMINI.md, portable prompts, and manifests.

Runtime and state

LangGraph, CrewAI, AutoGen, Semantic KernelRuntime adapters consume the neutral contract rather than provider-specific files.

Human-blocking gatesgate_waiting, approve/reject resume, and rejection cancellation before downstream work.

PostgreSQL-backed stateOrganizations, teams, projects, threads, runs, node outputs, approvals, errors, and final summaries.

Operations

FastAPI + provider routingModel prefixes route to Ollama, Gemini, Claude, or OpenAI without changing workflows.

Nunneri Graph StudioWorkflow phase config, traces, approvals, per-stage output, reruns, and cancellation.

RBAC and trust layerOIDC/local token auth, tenant-scoped permissions, security policy, governance, roadmap, citation, releases.

Object storage is optional; MinIO is never required core infrastructure.

AGPLv3 Community Edition with commercial licensing available.

Architecture and defensive publication are public-disclosure records.

Generated release artifacts keep docs, contracts, and adapters aligned.

Why Nunneri is different

Nunneri is not a single assistant prompt pack or a single orchestration framework. It is a neutral control plane that keeps intent, runtime behavior, approval rules, and evidence aligned across providers and runtimes.

Neutral source of truth

Canonical assets generate provider files, open-source manifests, runtime exports, and Graph Studio metadata from the same contract.

Human gates are real control flow

Approval phases pause execution and resume only through explicit approve or reject actions. Rejection cancels downstream implementation work.

State outside the model

Threads, runs, node outputs, approval decisions, errors, timing, and final summaries are persisted outside the LLM context window.

Tenant-aware execution

RBAC scopes dispatch, run history, phase configuration, and approvals across organization, team, project, thread, and run boundaries.

How it works

Three steps from zero to agentic workflows with human oversight and a full audit trail.

Clone and run

Clone the repo, start Ollama and PostgreSQL with Docker Compose. No cloud accounts, no API keys required to begin.

Launch Nunneri Graph Studio

Open the browser UI — select an agent, attach your project, and watch the LangGraph DAG execute node by node with live streaming and human approval gates.

Bring your models

Use any Ollama model locally. Add GEMINI_API_KEY or ANTHROPIC_API_KEY to switch to cloud LLMs with a single prefix — your choice, no lock-in.

Community Edition — AGPLv3

Everything you need,
self-hosted.

The Nunneri Community Edition — Nunneri Graph Studio, agentic workflows, human oversight gates, and multi-tenant RBAC — is available under AGPLv3. Run it on a laptop or any cloud, and review AGPLv3 obligations when modifying or providing network access to the software.

Nunneri Graph Studio UI with live DAG visualisation and node streaming
LangGraph stateful agentic workflows
Human approval gates — pause, review, resume or reject
PostgreSQL-backed run history, checkpoints, and error detail
Ollama local LLM support (Mistral, LLaMA 3, Phi, Qwen…)
Cloud LLMs via env var — Gemini, Claude — no lock-in
RBAC multi-tenant: org → team → project
Full run audit trail with per-node state
OIDC auth — Keycloak, Okta, Azure AD
Public source on GitHub — fork, extend, and contribute under AGPLv3

★ GitHub Repo Developer Setup → ⬡ Open Nunneri Graph Studio

Quick start

git clone https://github.com/suranku/nunneri-public-agentic-ai
cd nunneri-public-agentic-ai

# Start PostgreSQL
docker compose -f api/docker-compose.yml up -d

# Install Python dependencies
pip install -r api/requirements.txt

# Start the API server
uvicorn api.main:app --host 0.0.0.0 --port 8000 --reload

Open http://localhost:8000 for the portal or http://localhost:8000/ui for Nunneri Graph Studio.
Requires Ollama running locally with at least one model pulled (e.g. ollama pull mistral).

Full setup guide → Developer Guide · Nunneri Graph Studio User Guide

Commercial licensing

Nunneri follows an AGPLv3 Community Edition plus commercial-license model. The public repo remains available for community use, while commercial terms are available for organizations that need proprietary distribution, SaaS, OEM, managed-service, indemnity, or procurement coverage.

Community Edition

Use, modify, and distribute Nunneri under AGPLv3. Network use of modified versions may require corresponding source availability.

Read AGPLv3 →

Commercial License

Use Nunneri in closed-source, embedded, hosted, managed, or enterprise procurement scenarios under negotiated commercial terms.

Commercial terms →

Originators and Core Team

Nunneri was originated by Suranku and Yamini and is stewarded by the Nunneri Core Team.

Maintainers →

Architecture and defensive publication

Nunneri publishes implementation-level architecture material to document its runtime-neutral control plane, establish public disclosure, and make the system reproducible from the public repository.

Architecture

System components, canonical asset flow, runtime adapters, Graph Studio, RBAC, provider routing, and storage guidance.

Read architecture →

Defensive Publication

Implementation-level disclosure of the provider-neutral runtime contract, approval gates, state model, and tenant-scoped control plane.

Read disclosure →

Patent Notice

This publication documents the architecture for public-disclosure purposes. Do not use patent-pending language unless an application has actually been filed.

Runtime contract →

The Mission

Responsible AI and enterprise agentic workflows should not be the exclusive domain of large companies with large budgets. Nunneri makes them accessible, auditable, and self-hostable — for any team, any size, anywhere.

Run Anywhere

Fully air-gapped on a laptop with Ollama. No data leaves your machine unless you choose cloud models. Deploy on your hardware or any cloud with a single Docker Compose command.

No Lock-In

Standard PostgreSQL for state. Standard OIDC for auth. Standard FastAPI for the server. Switch models or switch clouds without rewriting workflows.

Responsible by Design

Human approval gates in every workflow. Full run audit trail with per-node state. Role-based access across org → team → project. Error detail stored and inspectable, never silently lost.