Just as developers use GitHub to collaborate on code, agents need Nebula to coordinate context.
Nebula is Git for agent context & tasks - a secure system of record with scopes, approvals, audit trails, and rollback.
MCP-first by design, CLI today, web UI & shared cloud next.
Nebula is perfect whether you’re building with multiple unrelated agents and need real collaboration (not copy-pasted prompts and messy markdown files), or if you just want agents to manage your notes and store context any approved agent can use.
git clone https://github.com/gravitrone/nebula-core.git
cd nebula-core
docker compose up -d
cd cli/src
go build -o build/nebula ./cmd/nebula
./build/nebula start # starts the API & MCP server
./build/nebula # open the CLIA one-line installation script is currently in development and will be available in the next release.
Nebula’s vision is to make agent context a reliable platform. Local-first stays core, but you’ll be able to push/commit your context to the cloud when you want. The mission is to replace messy markdown workflows with a real multi-agent communication protocol, with scopes, approvals, and auditability baked in, so shared context becomes as normal as shared code.
Nebula keeps your data as a connected graph (projects, context, jobs, logs, files).
Each agent sees only what its scopes allow.
Untrusted write requests go through approval first.
flowchart LR
A[Agents] --> B[Shared Graph]
B --> C{Scope Check}
C -->|Allowed| D[Show data]
C -->|Blocked| E[Hide data]
A -->|Write request| F{Trusted agent?}
F -->|Yes| G[Apply change]
F -->|No| H[Approval Queue]
H --> I[Human Review]
I -->|Approve| G
I -->|Reject| J[No change]
G --> B
G --> K[Audit History]
Simple rule - if scope says no, data stays hidden. if trust says no, writes need approval.
PRs are welcome.
Before submitting:
- Read the CLA and sign it by commenting: "I have read the CLA Document and I hereby sign the CLA"
- For major changes, open an issue first to align on scope
- Keep changes focused and include tests/docs when relevant
Why not just use Markdown or Notion?
They’re human-first. Nebula is agent-first - scopes, approvals, audit, rollback.
It lets unrelated agents/teams sync on shared context while a human keeps full control - like GitHub, but for agent memory.
You decide who can access what via scopes, so some context is shared and some stays private.
What’s the difference from a vector DB?
Nebula is a secure system-of-record with graph connections (entities, relationships, logs).
We also support embedding search with privacy scopes - local by default, and soon compatible with your favorite providers (e.g. OpenAI).
Is there a cloud version?
Coming. Web UI & shared cloud for seamless sync across your devices.
Can I rollback changes?
Yes - full audit log & one-click revert to any previous version.
Is my data private?
Yes. Data lives on your machine by default. Scopes control exactly who can read/write.
Can I import Markdown or existing notes?
JSON/CSV now. Full Markdown ingest is on the roadmap - but you can already ask any agent to do it.
What’s the CLI for?
It’s the human control panel - browse & add data, approve changes, run ops.
Agents connect through MCP/REST instead.