Use SynapCores as an Ollama server: chat, generate and embeddings API
SynapCores answers the Ollama API on the same port as the database. Models run inside the engine, so code written for Ollama works against SynapCores with a base URL and an API key, and the same models also power EMBED() and GENERATE() in SQL.
Install free, one command
curl -fsSL https://get.synapcores.com | SYNAPCORES_REF=docs-ollama-api shFree Community Edition. Linux (x86_64, aarch64) and macOS on Apple Silicon; Windows via WSL2. Docker and other options
You need a running engine (quickstart or Docker) and an API key in SYNAPCORES_API_KEY. The MCP guide shows how to create one with two commands; a session token from /v1/auth/login works too.
| Endpoint | What it does |
|---|---|
POST /api/chat | Chat completion, streamed as NDJSON or as one JSON object with "stream": false |
POST /api/generate | Text completion from a prompt |
POST /api/embed | Embeddings for one input or a list of inputs |
POST /api/embeddings | Single embedding, older request format ("prompt") |
GET /api/tags | Models available to your key |
POST /api/show | Model details: template, model info, capabilities |
GET /api/ps | Models loaded in memory |
GET /api/version | Version probe, no key needed |
List the models
curl -s localhost:8080/api/version | jq -c .
curl -s localhost:8080/api/tags -H "Authorization: Bearer $SYNAPCORES_API_KEY" \
| jq -r '.models[].name'{"version":"0.20.6"}
all-minilm:latest
d1:latest
qwen2.5-coder:7bYour list shows the models on your install. The installer offers a chat model sized to your RAM, and the engine downloads the all-minilm embedding model on first start.
Chat
MODEL=qwen2.5-coder:7b # any chat model from /api/tags
curl -s localhost:8080/api/chat -H "Authorization: Bearer $SYNAPCORES_API_KEY" \
-H 'Content-Type: application/json' \
-d "{\"model\":\"$MODEL\",\"stream\":false,\"messages\":[{\"role\":\"user\",\"content\":\"In one sentence: what is a vector database?\"}]}" \
| jq '{model, message, done}'{
"model": "qwen2.5-coder:7b",
"message": {
"role": "assistant",
"content": "A vector database is a type of database designed to store and manage data in the form of vectors, which can represent complex information such as text, images, or other high-dimensional data efficiently."
},
"done": true
}Leave out "stream":false to stream tokens as newline-delimited JSON, like Ollama:
curl -s -N localhost:8080/api/chat -H "Authorization: Bearer $SYNAPCORES_API_KEY" \
-H 'Content-Type: application/json' \
-d "{\"model\":\"$MODEL\",\"messages\":[{\"role\":\"user\",\"content\":\"Name three colors.\"}]}"{"model":"qwen2.5-coder:7b","created_at":"2026-11-18T16:02:01.857043000Z","done":false,"message":{"role":"assistant","content":"Sure"}}
{"model":"qwen2.5-coder:7b","created_at":"2026-11-18T16:02:01.907453000Z","done":false,"message":{"role":"assistant","content":"!"}}
{"model":"qwen2.5-coder:7b","created_at":"2026-11-18T16:02:01.976066000Z","done":false,"message":{"role":"assistant","content":" Here"}}
…Create embeddings
curl -s localhost:8080/api/embed -H "Authorization: Bearer $SYNAPCORES_API_KEY" \
-H 'Content-Type: application/json' \
-d '{"model":"all-minilm:latest","input":["SynapCores runs SQL and vector search in one engine","How do I reset my password?"]}' \
| jq -c '{model, count: (.embeddings | length), dims: (.embeddings[0] | length)}'{"model":"all-minilm:latest","count":2,"dims":384}With curl, send Content-Type: application/json on every POST, as in these examples. Ollama client libraries send it for you.
Point the official Ollama Python client at SynapCores
The ollama package reads OLLAMA_HOST and sends OLLAMA_API_KEY as a bearer token, so existing code needs no changes:
python3 -m venv .venv && . .venv/bin/activate
pip install ollama
export OLLAMA_HOST=http://localhost:8080
export OLLAMA_API_KEY=$SYNAPCORES_API_KEY
python demo.pyimport ollama
print([m.model for m in ollama.list().models])
reply = ollama.chat(
model="qwen2.5-coder:7b",
messages=[{"role": "user", "content": "In one sentence: what is a vector database?"}],
)
print(reply.message.content)
emb = ollama.embed(model="all-minilm:latest", input="SynapCores runs SQL and vector search in one engine")
print(len(emb.embeddings[0]), "dimensions")['all-minilm:latest', 'd1:latest', 'qwen2.5-coder:7b']
A vector database is a type of database designed to store and manage data in the form of vectors, which can represent complex information such as text, images, or other high-dimensional data efficiently.
384 dimensionsTo set the URL in code instead, use ollama.Client(host="http://localhost:8080", headers={"Authorization": "Bearer …"}).
Build on it
- Build a RAG app with LlamaIndex:
llama-index-llms-ollamatalks to the chat model on your engine with a base URL and your token, and retrieval runs in SQL. See the LlamaIndex guide. - Store your documents in the same engine and search them by meaning from Python, with no separate vector database.
- Keep retrieval in SQL:
EMBED()andCOSINE_SIMILARITY()in the quickstart use the same embedding model as/api/embed.
Next steps
Full reference
- SQL reference: Every statement and function: EMBED, VECTOR, COSINE_SIMILARITY, GENERATE, Cypher, AutoML, agent memory.
- API reference: The REST API. Your engine also serves its own OpenAPI schema at /v1/openapi.json.
- System requirements: Supported platforms, memory and disk sizing.