oh-my-ai

oh-my-ai is a Bun-native proxy runner for OpenAI-compatible and Gemini-compatible APIs.

It gives you one local endpoint with automatic API key rotation, provider-level access control, and a web admin panel for live configuration updates.

Why teams use it

• Automatic key rotation on retryable failures (default: 429)
• Multi-provider support (OpenAI, Gemini, Groq, OpenRouter, custom-compatible backends)
• Bun-first runner experience with bunx oh-my-ai
• Live admin panel to manage providers, keys, models, and rotation behavior
• Optional per-provider access keys for controlled usage
• In-memory request metrics and key/provider health visibility

Quick Start

Option A: Run instantly (recommended)

bunx oh-my-ai

On first run, if CONFIG_FILE does not exist, oh-my-ai creates a default config automatically.

Option B: Clone for local development

git clone https://github.com/ideo-mind/oh-my-ai.git
cd oh-my-ai
bun install
bun start

Open the admin panel at http://localhost:8990/admin.

Runtime Model

• Runtime: Bun ("engines": { "bun": ">=1.0.0" })
• Entrypoint: bunx oh-my-ai
• This project is a runner/server, not an SDK package.

Configuration

By default, config is read from:

• CONFIG_FILE (if set)
• otherwise ./config.toml in your current working directory

If the file is missing, a default config is created automatically.

Default generated config

[server]
port = 8990
adminPassword = "admin"

provider = {}

Recommended production baseline

[server]
port = 8990
adminPassword = "replace-this-with-a-strong-password"
# Optional
# logFilePath = "/var/log/oh-my-ai/proxy.log"

[provider.openai]
apiType = "openai"
apiKeys = ["sk-openai-key-1", "sk-openai-key-2"]
baseUrl = "https://api.openai.com/v1"
accessKey = "team-openai"
defaultModel = "gpt-4o-mini"
modelHistory = ["gpt-4o-mini", "gpt-4.1-mini"]
rotationStatusCodes = [429, 500, 502, 503]

[provider.gemini]
apiType = "gemini"
apiKeys = ["AIza...", "AIza..."]
baseUrl = "https://generativelanguage.googleapis.com/v1beta"
accessKey = "team-gemini"
defaultModel = "gemini-2.5-flash"
modelHistory = ["gemini-2.5-flash"]
rotationStatusCodes = [429, 500, 503]

Provider fields

• apiType: openai or gemini (lowercase recommended)
• apiKeys: array of API keys
• apiKeysDetailed: optional detailed key objects (value, label, tier)
• baseUrl: optional override (defaults are inferred by apiType)
• accessKey: optional per-provider gate key
• defaultModel: optional default model
• modelHistory: optional model history list
• rotationStatusCodes: optional HTTP codes that trigger key rotation

Environment Variables

• CONFIG_FILE: absolute or relative path to config file
• PORT: runtime override for server port
• ADMIN_PASSWORD: fallback only when config has no server.adminPassword

Port resolution order:

1. PORT
2. server.port from config
3. default 8990

Request Usage

OpenAI-compatible example

curl -X POST "http://localhost:8990/groq/chat/completions" \
  -H "Authorization: Bearer [STATUS_CODES:429][ACCESS_KEY:team-openai]" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-oss-120b",
    "messages": [
      {"role": "user", "content": "Say hello."}
    ]
  }'

Gemini-compatible example

curl -X POST "http://localhost:8990/gemini/models/gemini-2.5-flash:generateContent" \
  -H "x-goog-api-key: [STATUS_CODES:429][ACCESS_KEY:team-gemini]" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [
      {"parts": [{"text": "Say hello."}]}
    ]
  }'

If accessKey is not configured for a provider, omit [ACCESS_KEY:...].

Security Notes

• Change the default adminPassword before exposing the service.
• Keep CONFIG_FILE outside shared/public paths.
• Prefer unique accessKey values per provider in shared environments.

Releases

This repo uses Changesets.

• Create a changeset: bun run changeset
• Version locally: bun run version-packages
• Publish via workflow: push a tag like v1.0.2

CHANGELOG.md is generated from changesets during versioning.

Contributing

Contributions are welcome.

1. Open an issue describing the bug/feature.
2. Wait for alignment.
3. Submit a focused PR with clear rationale.

License

GNU GPL v3.0 or later. See LICENSE.