FULLY AIR-GAPPED
GI-1 runs entirely in your VPC or on-prem. Your repository never leaves your perimeter.
GÖDEL INTELLIGENCE
Coding agents,
aligned to your
repository.
GI-1 is a coding model and agent tuned to one enterprise
repository — its architecture, its conventions, its domain.
Model and harness keep evolving as your codebase grows.
SELF-EVOLUTION
It learns your repository, then rewrites itself.
A Gödel machine rewrites its own code the moment it can prove the change makes it better. GI-1 works the same way. Every task runs in an isolated sandbox with your repository mounted, and the outcome of each run — the diff, the tests, the review — becomes a verifiable training signal. Model and harness improve from every change you ship, so the intelligence compounds with your codebase instead of plateauing.
EVERY SURFACE
One aligned model, everywhere your engineers work.
GI-1 ships from the terminal, your desktop, the browser,
Slack, and your CI pipeline — always driving the same
repo-aligned model and the same agent harness.
- Godel CLI
- Desktop
- Web / Mobile
- Slack
- CI / Pipeline
REPO ALIGNMENT
It aligns to one repository.
GI-1 ingests your repository — architecture, dependency graph,
naming conventions, domain vocabulary, the history of what got
reviewed and merged — and builds a coding model and agent aligned
to that codebase. The code it writes reads like your team wrote it.
GI-1 CYBERDECK
Carry GI-1 in your pocket.
The GI-1 Cyberdeck is a handheld machine for running GI-1 away from your desk. A translucent deck with a mechanical keypad, an amber readout, and a copper rotary knob — dispatch agent runs, watch them work, and review the diff from anywhere.
Buy GI-1 Cyberdeck!ENTERPRISE
Built for the enterprise from day one.
Your repository is your most sensitive asset. GI-1 runs inside your perimeter, under your controls — alignment and self-evolution included.
ALIGNMENT ANALYTICS
See how aligned the model is to each repo — where it's strong, where it's still learning.
SAML / IDP INTEGRATION
Enterprise SSO with SAML 2.0, OIDC, and directory sync.
COST CONTROLS
Per-team and per-repo compute budgets with real-time tracking.
CENTRALIZED ORG CONFIG
Manage repositories, permissions, model routing, and policy from one pane.
OTEL NATIVE
First-class OpenTelemetry support out of the box.
IN PRODUCTION
Real repos.
Real diffs.
How teams ship with GI-1 on large enterprise codebases
Modularized a 1.2M-line payments monolith
▢ 38 prompts
⌗ 214 files
+9,180
−7,640
Untangled a 9-year-old billing service with no docs
▢ 51 prompts
⌗ 96 files
+2,410
~1,870
Strict null-safety across a legacy Java core
▢ 22 prompts
⌗ 308 files
+4,020
~3,110
◔ DEEP
Rewrote the CI pipeline as one reviewed PR
▢ 14 prompts
⌗ 47 files
+1,290
−2,640
◔ DEEP
RESEARCH
Notes from
the lab
GI-1 Enters Preview
GI-1 is open to design partners — repo-aligned coding intelligence for large enterprise codebases.
Aligning a Model to One Repository
The training method behind GI-1: how a base coding model is aligned to a single enterprise codebase.
Measuring Self-Evolution
A benchmark for whether the harness actually improves the longer it works a repository.
The Gödel Machine, In Practice
Turning Schmidhuber's self-rewriting machine into a production coding system.
The Open-Source Average
Why models trained on public code underperform on real enterprise repositories.
"We pointed GI-1 at a service two of our staff engineers barely understand. Within a week it was opening PRs that passed review on the first pass."
"Every other coding agent writes code that looks like the internet. GI-1 writes code that looks like our codebase."
"The thing that sold us: it got measurably better at our monorepo over the first month. Not the model in general — our repo specifically."
"It learned our internal RPC framework from the repository alone. No docs, no fine-tuning brief. It just read the code."
"We run it fully air-gapped. The repository never leaves our network and the model still improves. I didn't think that was possible."