---
title: Every model. One gateway. Yours
description: "Kaeso is an open-source AI gateway you run yourself, so applications talk to one API while you keep provider keys, routing and usage under your control."
url: https://kaeso.app
markdown: https://kaeso.app/index.md
language: en
alternates:
  de: https://kaeso.app/de/index.md
  zh-CN: https://kaeso.app/zh-CN/index.md
updated: 2026-11-13
kaeso_version: 0.6.0
index: https://kaeso.app/llms.txt
---

# Every model. One gateway. Yours

Kaeso is an open-source AI gateway you run yourself, so applications talk to one API while you keep provider keys, routing and usage under your control.

> Every page, article and docs page of Kaeso is listed in https://kaeso.app/llms.txt.

One API in front of every provider, run by you: 27 providers. Self-hosted · one Rust process · no enterprise edition.

[Install Kaeso](https://docs.kaeso.app/self-hosting/installation) · [Open the live demo](https://demo.kaeso.app)

## Own your keys. Own your limits. Own the record

- **Your keys never leave the gateway**: Provider keys are encrypted with AES-256-GCM under a master key you hold. Apps get Kaeso keys instead, stored only as hashes.
- **Spend stops where you say**: Budgets and rate limits per project, member, service account and key. A hard budget refuses before any provider is called.
- **Know who asked. Know what it cost**: Every request lands in your own log with its key, member, model, tokens, time and its cost at the prices in force.

## Command every call

Organizations, projects, keys, limits, routes, request logs and a playground, in one console beside the gateway. This is the console itself, on sample data. Click through it.

[Open the live demo](https://demo.kaeso.app)

[Kaeso 0.6.0](https://kaeso.app/changelog/0.6.0.md) (10 November 2026): Chat, images, speech and video through one gateway.

## No model call without a key, a limit and a log

One Rust process between your apps and every provider. It checks the key and the limits, calls the provider with a key only it holds, and writes down what each request used and cost.

[How a request travels](https://kaeso.app/gateway.md)

- [Request logs](https://docs.kaeso.app/concepts/request-flow): Every call, its time and its price
- [Models](https://docs.kaeso.app/concepts/alias-routes): One catalogue, with aliases and routes
- [Budgets](https://docs.kaeso.app/console/limits-and-budgets): A hard one refuses before you pay
- [27 providers](https://kaeso.app/providers.md): Built in, each called in its own format: OpenAI, Anthropic, Baseten, Cerebras, Cohere, DeepSeek, ElevenLabs, Fireworks AI, Gemini Enterprise Agent Platform, Google AI Studio, Groq, Hugging Face, MiniMax, Mistral, Moonshot, Nebius Token Factory, NVIDIA, Ollama Cloud, OpenRouter, QwenCloud, SambaNova, Scaleway, SiliconFlow, SpaceX, Together AI, Vercel and Z.ai
- [Usage](https://demo.kaeso.app): Thirty days of the sample organization
- [Single sign-on](https://docs.kaeso.app/self-hosting/sign-in): Your OpenID Connect provider, roles, two-factor
- [MCP portal](https://docs.kaeso.app/concepts/mcp-portal): A changed tool waits for review

## Every request answers to you

Apps keep their SDK and change two settings. From there, every call goes through the same eight checks, in the same order, before a provider ever sees it.

- **Route**: An alias picks the model: fallbacks when a provider fails, conditions on headers, formats and tokens, weighted splits.
- **Limit**: Rates and budgets per project, member, service account and key. A hard budget refuses before the provider is called.
- **Record**: Every request lands in the log with its model, tokens, time and cost at the prices in force.

**Nothing reaches a provider unchecked**

### Eight steps, in order

1. **Key**: The key decides the project and who it acts for. Kaeso stores it only as a hash.
2. **Decode**: The body is read in the app's own format. Anything Kaeso would never send on is refused here.
3. **Model**: The name resolves in the project's catalogue. An alias's route picks the model and its fallbacks.
4. **Limits**: Every rate and budget of the key, the member and the project, checked at once.
5. **Provider key**: The project's key for the provider, decrypted from AES-256-GCM under your master key.
6. **Encode and send**: The request is written in the provider's own format, and sent.
7. **Reply**: Decoded, and encoded back into the app's format, event by event for a stream.
8. **Log**: Usage, the cost at the prices in force, and the request log, after the reply has gone out.

[The request path](https://kaeso.app/gateway.md)

### Built to run yourself

- **Single sign-on**: Per organization, through its own OpenID Connect provider, with two-factor and roles. (OIDC)
- **Provider keys**: Encrypted under a master key you hold. Kaeso's own keys are stored only as hashes. (AES-256-GCM)
- **Budgets**: A hard budget refuses before the provider is called; a soft one sends an alert. (Hard, soft)
- **MCP portal**: One portal per project. A tool whose definition changes waits for an admin. (Held for review)

## One door. Every model

Apps keep the OpenAI, Anthropic and Google SDKs and change two settings: the base URL and the key. Kaeso calls each of the 26 chat providers in its own format.

[Every provider](https://kaeso.app/providers.md)

- Providers built in: 27
- Formats apps speak: 4
- Checks per request: 8
- Rust process: 1

Every chat format an app can speak, against the 26 chat providers built into Kaeso: filled where Kaeso calls the provider in that format, open where it translates. Point at a cell, touch one, or tab to one.

- OpenAI: Responses, Chat Completions
- Anthropic: Messages
- Baseten: Chat Completions
- Cerebras: Chat Completions
- Cohere: Chat Completions
- DeepSeek: Chat Completions
- Fireworks AI: Chat Completions, Messages
- Gemini Enterprise Agent Platform: Gemini API, Messages, Chat Completions
- Google AI Studio: Gemini API
- Groq: Chat Completions
- Hugging Face: Chat Completions
- MiniMax: Messages
- Mistral: Chat Completions
- Moonshot: Chat Completions
- Nebius Token Factory: Chat Completions
- NVIDIA: Chat Completions
- Ollama Cloud: Chat Completions
- OpenRouter: Chat Completions
- QwenCloud: Chat Completions
- SambaNova: Chat Completions, Messages
- Scaleway: Chat Completions
- SiliconFlow: Chat Completions
- SpaceX: Responses, Chat Completions
- Together AI: Chat Completions
- Vercel: Chat Completions, Responses, Messages
- Z.ai: Messages

## Not a service. Software you run

Caddy, the gateway, the console, PostgreSQL, ClickHouse and Valkey, with Docker Compose. Or a template for Render, Railway, DigitalOcean, Fly.io, AWS, Azure, Google Cloud, Coolify, Dokploy and CapRover.

[Read the installation guide](https://docs.kaeso.app/self-hosting/installation)

```bash
curl -fsSL https://kaeso.app/install.sh | sh
```

install.sh, on a fresh server:

```
$ curl -fsSL https://kaeso.app/install.sh | sh

==> Installing Kaeso in /home/ada/kaeso

==> Downloading Kaeso's images, about 2 GB; this takes a minute or two

==> Creating the master key

==> Starting Kaeso

==> Kaeso is running. Open this link to create the first user:

https://localhost/setup#token=…
```

## How it compares

Checked on 6 November 2026 against each project's repository and documentation, where the others do more too.

[See the sources in the README](https://github.com/logicplanes/kaeso#how-it-compares)

| Feature | Kaeso | LiteLLM | Bifrost | Portkey Gateway |
| --- | --- | --- | --- | --- |
| Gateway written in | Rust | Python | Go | TypeScript |
| Licence | MIT | MIT, except enterprise/ | Apache 2.0 | MIT |
| Single sign-on when self-hosted for free | Yes, OpenID Connect | Up to 5 users | Enterprise | Enterprise |
| Budgets when self-hosted for free | Yes | Yes | Yes | Enterprise and some Pro plans |
| Web console when self-hosted | Yes | Yes | Yes | Request logs only |
| MCP portal for your own MCP servers | Yes, per project | Yes | Yes | Yes |

## News and articles

- [Bring your own key to Linear's coding agent, through Kaeso](https://kaeso.app/articles/linear-byok-through-kaeso.md) (13 November 2026): Linear's cloud coding agent now takes your own key, and any gateway that speaks Anthropic Messages or OpenAI Responses. Point it at Kaeso to put its sessions under your own keys, limits, routes and request logs. Step by step.
- [Kaeso 0.6: one gateway for chat and media, with alias routes and an MCP portal](https://kaeso.app/articles/kaeso-0-6.md) (11 November 2026): Kaeso 0.6.0 is out. Apps make images, speech and video through the same gateway as chat, Gemini apps connect under /v1beta, aliases become routes with fallbacks and splits, and every project gets an MCP portal. Here is what changed and how to update.
- [Introducing the MCP portal](https://kaeso.app/articles/mcp-portal.md) (11 November 2026): Every Kaeso project now has an MCP portal. Connect your organization's MCP servers once, choose which tools each project offers, and let claude.ai, ChatGPT, Cursor and your own agents call them with Kaeso's sign-in, rights, limits and logs.

[All articles](https://kaeso.app/articles.md)

## Own the record

Free. All of it. Single sign-on, budgets and the console are in the MIT-licensed code. There is no enterprise edition.

[Install Kaeso](https://docs.kaeso.app/self-hosting/installation) · [Open the live demo](https://demo.kaeso.app)
