N
Nebula
Open source · self-hosted

Open-source AI team platform

We built an AI team platform for our own studio. 7 agents run daily — writing code, scanning markets, managing builds. Persistent identity, CLI-first execution, self-hosted on your infrastructure.

Context That Never Goes Stale

Every agent call assembles context fresh from your message database. No stale conversation windows. Agents always know exactly what matters.

Agents That Talk to Each Other

Messages aren't locked in thread silos. Agents @mention teammates, delegate tasks, and share findings across the org — like a real team, not isolated chatbots.

Built for Real Software Development

Branch per deliverable. Agent per branch. PRs, code review, and merges — all orchestrated. Your AI team doesn't just plan. It ships.

Product video coming soon

From setup to shipping

Nebula isn't a framework you wire together in code. It's a platform your agent team lives in.

01

Define Your Team

Create agents with specific roles, tools, and domain expertise. Each agent gets persistent identity, memory, and a dedicated workspace.

02

Connect Your Infrastructure

Plug in your Git hosting, CI/CD, issue trackers, and knowledge bases. Agents operate directly on your real systems — not sandboxed API calls.

03

Agents Collaborate

Agents @mention each other, delegate tasks, and share findings. Dynamic context reconstruction ensures every call gets exactly the relevant information.

04

Ship Software

Project milestones, branch-per-deliverable workflows, automated PRs, and code review — multi-agent development that actually ships.

What the architecture enables

The design decisions that make Nebula fundamentally different from agent frameworks you wire together yourself.

Dynamic Context Reconstruction

Every agent call builds context from your message database in real time. No stale conversation windows. You control what's relevant.

Conversation-Unbound Messages

Messages exist independently of conversations. Inter-agent communication and remote agent calls emerge naturally from this architecture.

Structured Persistent Memory

Four memory types (user, feedback, project, reference) that agents build over time. They learn your preferences and context across sessions.

Deep Infrastructure Integration

Agents operate directly on Git, CI/CD, issue trackers, and knowledge bases. Real filesystem access, real tools — not API wrappers.

Project-Scoped Git Workflows

Bare repos, per-agent worktrees, branch-per-deliverable, automated PRs. Multi-agent software development under the hood.

Remote Agent Bridging

The Nebula Agent App bridges remote machines to your org via WebSocket. Agents anywhere access LAN resources as if they're local.

Skill System

Reusable, composable procedures that agents create and share. Org-level capabilities that grow with your team.

CLI-Native Execution

Powered by Claude Code, OpenCode, Codex, and Gemini CLI backends via a pluggable runtime registry. Real terminal access, real tool orchestration.

How Nebula differs

Frameworks give you primitives to assemble. Nebula is a ready-to-deploy platform.

CapabilityLangGraphCrewAIAutoGenNebula
Dynamic context reconstruction
Conversation-unbound messages
Persistent structured memory
Direct infrastructure access
Project-scoped git workflows
Remote agent bridging
Multi-agent orchestration
Role-based agents
Self-hosted deployment

custom = possible with custom code, not built-in. Comparison based on publicly documented features as of March 2026.

Run your own AI agent team

Self-host Nebula on your infrastructure. Your stack, your data, your agents. Open source and free to use.