- 100% open source, serves AI models via a REST interface
- Runs open source models locally (using Ollama), and on the cloud (Fly + Supabase)
- Wraps popular closed models
- Stores queries/responses in Postgres for training
- Secure by default
- Is as simple as possible
ollama run llama2supabase start
Then make a request to the proxy function:
curl -i --location --request POST 'http://127.0.0.1:54321/functions/v1/proxy' \
--header 'Authorization: Bearer eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJpc3MiOiJzdXBhYmFzZS1kZW1vIiwicm9sZSI6ImFub24iLCJleHAiOjE5ODM4MTI5OTZ9.CRXP1A7WOeoJeXxjNni43kdQwgnWNReilDMblYTn_I0' \
--header 'Content-Type: application/json' \
--data '{
"model": "llama2",
"messages": [
{
"role": "system",
"content": "You are a helpful assistant."
},
{
"role": "user",
"content": "Hello!"
}
]
}'This will store the data in the database for later training.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent