Best experienced on desktop

This demo uses interactive visualizations designed for larger screens. Open it on your laptop or desktop for the full experience.

Mobile version coming soon
GitLab Orbit  ·  Internal Preview  ·  18.11

Your AI agents
are flying blind.

At enterprise scale, codebases span hundreds of repositories, millions of lines of code, and thousands of pipelines. Your AI sees a few hundred files. Orbit gives it the full map.

See how it works Get in touch
↓ Scroll to explore
session session-mgr token token-store auth-db auth-db notify notify-svc pay payments-core ci-1 ci-deploy-1 ci-2 ci-deploy-2 srch search-svc plt @platform ? ? ? ? ? ? ? ? ? AI CONTEXT WINDOW auth- service billing- core api- gateway ≈ 3 of 12 services visible to the AI
WITHOUT GKG WITH ORBIT MONOLITH auth api db utils ? ? ? ? MICROSERVICES auth-svc billing api-gw notify-svc ? ? ? ? Dependencies unknown · AI guesses MONOLITH auth api db utils calls reads uses writes MICROSERVICES auth-svc billing api-gw notify-svc called by calls events routes to All relationships mapped · exact
The Problem
The problem

Your AI agents
are flying blind.

Every AI agent operates on a context window. That window holds a few hundred files. Enterprise codebases have hundreds of thousands. The cross-repo dependency chains, ownership graphs, and pipeline relationships that matter most fit in none of it.

Inference, not fact

AI reads whatever fits in its context window and guesses the rest. In a 500,000-line codebase, the answer depends on which 200 files it happened to see. Not what is actually true.

Context window limits

A typical enterprise codebase spans 100,000+ files across dozens of repos. The context window fits roughly 200. The dependencies that will break your next deploy are in the other 99,800.

Cross-repo blindness

Blast radius, ownership, pipeline tracing across hundreds of services and thousands of pipelines require traversing relationships that span the entire org. That is only possible with a graph.
Every architecture

Monolith or microservices —
same blind spot.

It does not matter how your codebase is structured. At enterprise scale, the import chains, the callers, the owners, and the blast radius span hundreds of repos and thousands of files. Without a graph, all of it is unknown.

Monolith

500,000 files · AI sees ~200

Internal module dependencies are opaque. Changing AuthService might break 147 functions across 5 modules. AI guesses 3 to 5. It cannot know without traversing the full call graph.

Microservices

200 services · 0 queryable relationships

Cross-service dependencies live across 200 repos. auth-service is called by 23 services you don't know about. AI can't see any of it.

The solution

Orbit maps everything.

Code. Pipelines. CI/CD metadata. Team ownership. import trees. Cross-repo dependency chains across hundreds of repositories and millions of definitions. All indexed into a structured, queryable knowledge graph, updated in near real-time via CDC. This is the infrastructure that makes AI reliable at enterprise scale.

9
languages
<300ms
query latency
Indexed

Ruby, Java, Kotlin, Python, TypeScript, JavaScript, Rust, Go, C# — full cross-file refs.

def authenticate() → every caller, in every repo, at any depth.

Real-time

CDC pipeline: always current

Not a snapshot. Every git push triggers re-indexing. The graph reflects what's actually deployed.

Real-world use cases

The questions that
actually matter.

Blast radius. Pipeline inheritance. Service dependency mapping. Disaster recovery. These are the questions enterprise teams ask on day one. Orbit answers all of them in under 300ms.

Change impact

"What breaks if I change this service?"

Every caller, every downstream dependency, every pipeline, every owner, across every repo, in one query. No manual grep. No guessing.

Pipeline standards

"Which teams are out of compliance with our security pipeline?"

Map include chains, find who's drifted from the golden pipeline, surface missing security scans across every namespace.

Disaster recovery

"Map every service before we migrate."

Full dependency graph before an infrastructure migration. Know exactly what touches what. No surprises on cutover day.

See blast radius in action ↓
Integrations

Works with the agents
you already use.

Orbit plugs natively into Duo and via open MCP interfaces to Claude Code, Codex, and any agent that speaks the protocol.

Native · Zero-rated

GitLab Duo

Orbit is the context engine powering Duo's answers. Every query runs against the knowledge graph automatically. DAP queries are zero-rated.

MCP · 2 tools

Claude Code & Codex

Two MCP tools give complete graph access: query_graph and get_graph_schema. Connect in seconds. Billed via GitLab Credits.

REST · CLI

Any agent or tool

Full REST API at /api/v4/orbit/query and glab orbit query CLI for direct integration or custom tooling.

Duo Agent Platform + Orbit

Graph intelligence.
Native in GitLab Duo Agent Platform.

Orbit knows what's connected. Duo acts on it: creating issues, notifying teams, enforcing governance, and automating workflows without leaving GitLab.

Natural language on your graph

Ask in plain English. Orbit returns exact, cross-repo data. Duo contextualizes, explains, and suggests next steps. No graph syntax required.

Agentic follow-through

Duo uses Orbit findings to create issues, notify owners, draft runbooks, and trigger pipelines. One question, complete resolution.

Executive-ready insights

CI/CD health, delivery metrics, cross-team dependencies. Generate exec-reportable outcomes from your SDLC graph on demand.

GitLab Duo Chat
Claude Opus 4.6 · Anthropic ▾ Agentic
Let's work through this together...
Real-world use cases

What your team will actually ask.

Every question that used to require hours of manual repo-spelunking now takes under 300ms. These are the use cases resonating with enterprise teams today.

01  ·  Change Impact

What breaks if I change this service?

Full blast radius: every direct caller, downstream dependency, pipeline, and owner across every repo in your org, in a single query_graph call.

02  ·  Org Intelligence

How do our 200 services depend on each other?

Cross-repo dependency mapping at org scale. Discover which services call which, which repos share infrastructure, and which teams own what. No cloning required.

03  ·  Pipeline Standards

Which teams have drifted from our security pipeline standards?

Map include chains across thousands of projects. Find who's missing required security scans, who's still on a deprecated template, and who's diverged from the golden pipeline.

04  ·  Disaster Recovery

Map everything before we migrate.

Generate a complete service dependency graph before an infrastructure migration. Know exactly what touches what so there are no surprises when you flip the switch.

05  ·  Onboarding

Help me understand this codebase in a day.

New engineers get an instant, queryable map of the entire org: who owns what, how services connect, what's safe to change. Onboarding in days, not weeks.

06  ·  Vulnerability Tracing

Where is this CVE exposed across our repos?

Trace a vulnerable function, library, or pattern across every repo and namespace. Get a full impact list with owners in seconds. No manual search required.

See it in action

The same question.
Two very different answers.

Watch what Claude gets without Orbit, and what it gets when it can call query_graph on your knowledge graph.

claude  ·  ~/projects/gitlab
How it works

Index. Connect. Query.

Group owners enable indexing per namespace. Agents connect in minutes. Answers are instant, grounded, and cross-repo.

01 Index
NamespaceIndexing
◈ acme-corp Top-level
try it
└─ ◻ acme-corp / platform
└─ ◻ acme-corp / services
└─ ◻ acme-corp / infra
No namespaces indexing — toggle one to begin

Group owners enable indexing from GitLab admin settings. No code changes needed.

›
02 Connect
GitLab Duo Chat
GitLab Duo Agent Platform
Ask Orbit anything about your SDLC.
What's the blast radius of authenticate()?
Which pipelines depend on billing-core?
Claude Opus 4.6 ▾ Agentic

Zero-rated. Included in Duo Developer. No configuration needed.

claude  ·  with Orbit MCP
# ~/.claude/mcp.json
{
  "orbit": {
    "type": "stdio",
    "command": "glab",
    "args": ["orbit", "mcp"]
  }
}
 
# Tools available after connection:
✓ query_graph  ·  graph traversal
✓ get_graph_schema  ·  schema discovery

Works with Claude Code, Codex, and any MCP client. Billed via GitLab Credits.

›
03 Query
❴ "What would break if I change authenticate()?"
✓Callers (direct)api-gateway, billing-core, session-mgr +20
✓Downstream3 services at depth=2
✓Repos7 repos · 3 namespaces
✓Pipelinesci-deploy-1, ci-deploy-2
✓Owner@platform-team
187ms · deterministic · cross-repo
Enable indexing to unlock answers

Deterministic, cross-repo, grounded. No hallucination on structure.

Technical reality

The hidden cost of flying blind.

Without a graph, every AI query carries overhead that compounds at enterprise scale: API sprawl, auth on every hop, and context windows stuffed with irrelevant code.

Without Orbit

Excessive overhead

  • N API calls per query — one per repo, one per file type, one per service. Each hop adds latency and error surface.
  • Auth on every hop — MCP auth, repo auth, service auth. Each adds round-trip latency and token management complexity.
  • Context window bloat — thousands of irrelevant lines loaded to find one relationship. Tokens wasted on noise.
  • LLM inference on structure — the model guesses missing edges. Error compounds with every hop. The answer drifts from truth.
Minutes to find a blast radius across enterprise repos
With Orbit

Deterministic traversal

  • 1 call: query_graph — single MCP tool, all data in one structured response regardless of depth.
  • Graph-level auth — one namespace-scoped token. Zero per-service overhead. No auth sprawl.
  • Exact nodes and edges only — Orbit returns only what was requested. No context waste, no irrelevant files.
  • Indexed facts, not inference — traversal follows real, pre-indexed edges. The same query returns the same answer every time.
<300ms same exact answer, every time, at any depth
Integrations

Works with the agents you already use.

Orbit plugs natively into Duo and via open MCP interfaces to Claude, Codex, and any agent that speaks the protocol.

Native  ·  Zero-rated

GitLab Duo

Orbit is the context engine powering Duo's answers. Every query runs against the knowledge graph automatically. No configuration needed.

Duo Agent PlatformOrbit Agent
MCP  ·  2 tools

Claude Code & Codex

External AI agents connect via MCP. query_graph and get_graph_schema give complete access. Billed via GitLab Credits.

query_graphget_graph_schema
REST  ·  CLI

Any Agent or Tool

Full REST API and glab orbit query CLI for teams that want direct integration, custom dashboards, or pipeline tooling.

/api/v4/orbit/queryglab orbit
Language support

Indexing the stack you actually run.

9 languages indexed. Full cross-file reference graphs across your entire stack. Coverage expanding every milestone.

Full Cross-file references, call graphs, and definitions at any depth across all repos
Ruby
Java
Kotlin
Python
TypeScript
JavaScript
Rust
Go New
C# New
Coming soon
C
C++
On the roadmap
Swift
COBOL
Terraform / HCL
Release roadmap

From beta to GA.

Now in internal beta. Here is what is coming and when.

Now
Apr 2026
18.11
Internal beta. Design partner onboarding begins on GitLab.com SaaS.
Next
May 21, 2026
19.0
Public beta on GitLab.com. Available to all Premium and Ultimate customers.
Planned
Q2 FY27
Self-Managed
Self-managed deployments available. Customer-provisioned K8s with OAK or Helm.
Planned
Q3 FY27
GitLab Dedicated
GitLab-managed single-tenant deployments. Full isolation, no customer infra required.
Roadmap
Post Q3 FY27
GA
General Availability across all deployment types. SaaS, Self-Managed, and Dedicated all supported.
Deployment

How to access Orbit.

Orbit runs on GitLab.com today. Self-managed ships Q2 FY27. GitLab Dedicated follows in Q3 FY27.

Open Beta · 19.0

GitLab.com (SaaS)

Cloud-hosted. No infrastructure to manage. Requires Premium or Ultimate.

1

Get access

Join the design partner program. Orbit enables indexing on your namespace via a beta flag - no setup needed on your side.

2

Use Duo (native, zero-rated)

Orbit powers the Orbit agent in Duo Agent Platform automatically. Ask questions - Duo answers using your knowledge graph. DAP queries are zero-rated.

3

Connect external agents via MCP

Add Orbit to Claude Code, Codex, or any MCP-compatible agent in seconds. Billed via GitLab Credits.

External agent setup
Claude Code
→
query_graph
→
Orbit API
Codex
→
get_graph_schema
→
Orbit API
/api/v4/orbit/query
·
glab orbit query
Q2 FY27

Self-Managed

Your instance. Your data. GKG components run on a Kubernetes cluster alongside your GitLab install.

OAK
Omnibus GitLab + separate K8s cluster for GKG
Helm
Cloud-native GitLab - extend existing K8s with GKG charts
Data pipeline
GitLab
→
Siphon
→
NATS
→
GKG Indexer
→
ClickHouse
Your Agent
→
Orbit API
→
ClickHouse
✓

Same MCP surface

query_graph and get_graph_schema - no agent changes required.

✓

Data never leaves your environment

Cluster can be network-separated from GitLab. TLS supported for split-network deployments.

K8s

Customer-provisioned

GKE, EKS, AKS, or on-prem. ClickHouse is separate (self-hosted or ClickHouse Cloud). GitLab engineering onboards hands-on.

Shipping Q2 FY27
Q3 FY27

GitLab Dedicated

Single-tenant. GitLab-managed. No infrastructure to provision - full isolation without the ops burden.

Architecture
GKG runs inside your Dedicated tenant. GitLab manages all underlying infrastructure.
✓

Same MCP surface

query_graph and get_graph_schema - no agent changes required.

✓

Full tenant isolation

Your data never touches shared infrastructure. Dedicated AWS account, no multi-tenancy.

✓

No ClickHouse procurement

GitLab manages ClickHouse within your tenant. No separate vendor relationship required.

Shipping Q3 FY27
Design partner program

Stop letting your AI guess.

Orbit is invite-only on GitLab.com today. Design partners get a direct line to the team and the ability to shape what gets built.

Get in touch