Ship AI code. Keep control.
One gateway for Claude, GPT, Gemini, NVIDIA and Ollama — with budgets, model policy, secret filtering and a full audit trail of who changed what and why. Self-hosted, on your infrastructure, with your own provider keys.
Runs on a single EC2 box. No code leaves your infrastructure except to the providers you approve.
A cheap model narrows the problem.
The expensive one solves it.
Most AI coding tools throw your whole repository at a frontier model and bill you for it. Gatway does the thinking about what to send first — so you pay for the code that matters.
Analyze & optimize
A small, cheap model rewrites the prompt for clarity and asks a clarifying question if the request is ambiguous — before a single expensive token is spent.
Locate the code
It identifies the exact files and functions involved. Your editor slices out just those symbols — not the first 5,000 characters of five files.
Send only what matters
The optimized prompt plus the targeted code goes to the strong model. Typical requests drop from ~34k characters to ~3k.
Apply & record
Changes come back as a reviewable diff. Approve, and the edit is applied and written to the audit trail with the prompt that caused it.
Everything your team needs.
Everything your CTO asks for.
Inline autocomplete
Ghost-text completions as you type, routed to a fast model of your choosing — including a local one, so keystrokes never leave the machine.
Edit with diffs
The model proposes changes; you see a native diff and approve. Nothing is written to disk without a human clicking Apply.
AI code review
Review any diff or pull request against your project rules. Structured findings with severity, file, line and a concrete fix. Works from CI too.
Governed agents
Declarative agents that explore the codebase and propose multi-file changes. Tool allowlists are enforced by the gateway — a read-only agent truly cannot write.
Secret & PII filter
Outbound prompts are scanned for keys, tokens and PII. Log, redact or block them — per organization policy, before anything reaches a provider.
Change audit trail
Every prompt and every applied change, with the full diff, attributed to a developer and a project. Leads can answer "who changed this, and why?".
Cost controls
Semantic response cache, context optimizer, focused context and cheap-vs-strong routing — plus hard monthly budgets per org, project and user.
Provider failover
Rate limit or outage on one provider? Requests fail over to the next model in your chain automatically. Your team never sees it.
Bring your own keys
Your Anthropic, OpenAI, Gemini and NVIDIA keys, encrypted at rest. Your vector database. Your server. We never sit in the middle.
Built for teams that can't just paste code into a SaaS.
Gatway runs inside your perimeter. Approve the providers, set the policy, and get the evidence trail your security review will ask for.
Pay for infrastructure, not per seat.
Gatway sits between your team and the model providers. You pay the providers directly — we don't mark up your tokens.
Self-hosted
Run the whole platform on your own server. Ideal for teams that want control today.
- Unlimited developers
- All providers, your own keys
- VS Code extension, agents, code review
- Cost dashboard & change audit
- Community support
Enterprise
For organizations that need identity integration, deployment support and an SLA.
- Everything in Self-hosted
- SSO / SAML and directory sync
- Deployment & migration support
- Custom agents and policy design
- Priority support with an SLA
See it on your codebase.
A 30-minute walkthrough: connect your providers, run an agent on a real repository, and see the audit trail and cost breakdown it produces.
Or email hello@gatway.ai