fleet-embeddings（舰队嵌入）

Fleet Embeddings

You're helping someone generate embeddings — converting text into vectors for semantic search, RAG pipelines, duplicate detection, or recommendation systems. Instead of hitting one Ollama instance, the fleet distributes embedding requests across all available nodes automatically.

Why fleet embeddings matter

Building a RAG knowledge base means embedding thousands of document chunks. On a single machine, embedding 10,000 chunks takes significant time and blocks LLM inference. With fleet routing, embedding requests spread across nodes — the machine that's least busy handles each batch, and LLM inference continues uninterrupted on other nodes.

Same Ollama embedding models you already know. Same API. Just faster because the fleet parallelizes it.

Get started

pip install ollama-herd
herd                        # start the router (port 11435)
herd-node                   # start on each device
ollama pull nomic-embed-text  # pull an embedding model

No feature toggle needed — embeddings route through Ollama automatically.

Package: ollama-herd | Repo: github.com/geeks-accelerator/ollama-herd

Generate embeddings

Ollama format (curl)

curl http://localhost:11435/api/embeddings -d '{
  "model": "nomic-embed-text",
  "prompt": "The fleet manages all inference routing"
}'

OpenAI SDK (Python)

from openai import OpenAI

client = OpenAI(base_url="http://localhost:11435/v1", api_key="not-needed")

response = client.embeddings.create(
    model="nomic-embed-text",
    input="The fleet manages all inference routing",
)
vector = response.data[0].embedding
print(f"Dimensions: {len(vector)}")

Python (httpx)

import httpx

def embed(text, model="nomic-embed-text"):
    resp = httpx.post(
        "http://localhost:11435/api/embeddings",
        json={"model": model, "prompt": text},
        timeout=30.0,
    )
    resp.raise_for_status()
    return resp.json()["embedding"]

vector = embed("search query here")

Batch embedding for RAG

import httpx

def embed_batch(texts, model="nomic-embed-text"):
    """Embed a list of texts. Fleet distributes across nodes."""
    vectors = []
    for text in texts:
        resp = httpx.post(
            "http://localhost:11435/api/embeddings",
            json={"model": model, "prompt": text},
            timeout=30.0,
        )
        resp.raise_for_status()
        vectors.append(resp.json()["embedding"])
    return vectors

# Embed document chunks for RAG
chunks = [
    "Introduction to fleet management...",
    "The scoring engine uses 7 signals...",
    "Context protection prevents model reloads...",
]
vectors = embed_batch(chunks)
print(f"Embedded {len(vectors)} chunks, {len(vectors[0])} dimensions each")

Available embedding models

Check what's available:

curl -s http://localhost:11435/api/tags | python3 -c "
import json, sys
for m in json.load(sys.stdin)['models']:
    if 'embed' in m['name'].lower() or 'nomic' in m['name'].lower():
        print(f'  {m[\"name\"]}')"

Common models: nomic-embed-text, mxbai-embed-large, all-minilm, snowflake-arctic-embed.

Pull a model if needed:

curl -X POST http://localhost:11435/dashboard/api/pull \
  -H "Content-Type: application/json" \
  -d '{"model": "nomic-embed-text", "node_id": "your-node-id"}'

Usage analytics

Tag embedding requests to track per-project usage:

resp = httpx.post(
    "http://localhost:11435/api/embeddings",
    json={
        "model": "nomic-embed-text",
        "prompt": text,
        "metadata": {"tags": ["my-rag-pipeline", "indexing"]},
    },
)

Also available on this fleet

LLM inference

curl http://localhost:11435/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-oss:120b","messages":[{"role":"user","content":"Hello"}]}'

Drop-in OpenAI SDK compatible. 7-signal scoring routes to the optimal node.

Image generation

curl -o image.png http://localhost:11435/api/generate-image \
  -H "Content-Type: application/json" \
  -d '{"model":"z-image-turbo","prompt":"a sunset","width":1024,"height":1024,"steps":4}'

Requires FLEET_IMAGE_GENERATION=true. Uses mflux (MLX-native Flux).

Speech-to-text

curl -s http://localhost:11435/api/transcribe \
  -F "audio=@recording.wav" | python3 -m json.tool

Requires FLEET_TRANSCRIPTION=true. Uses Qwen3-ASR.

Monitoring

# Fleet health and model recommendations
curl -s http://localhost:11435/dashboard/api/health | python3 -m json.tool

# Per-app usage (see which projects use the most tokens)
curl -s http://localhost:11435/dashboard/api/apps | python3 -m json.tool

Dashboard at http://localhost:11435/dashboard — embedding requests flow through the same queues as LLM requests.

Full documentation

Agent Setup Guide — complete reference for all 4 model types.

Request Tagging Guide — tag requests for per-project analytics.

Guardrails

• Never delete or modify files in ~/.fleet-manager/.
• Never pull or delete models without user confirmation.
• If embedding model not available, suggest: ollama pull nomic-embed-text.
• If router not running, suggest: herd or uv run herd.