Quick start
Get a working Veda cluster and run your first vector search in a few minutes. You need Docker with Compose and git — there is no published Veda image yet, so Compose builds the server from source in this repo.
Get the repository
Compose builds Veda from this repo, so clone it first — docker compose has nothing to read otherwise.
git clone https://github.com/kshitij2605/Veda.git
cd VedaStart the core services
# Copy and configure environment
cp .env.example .env
# Core services (Veda + MinIO + NATS)
docker compose upThe first up compiles the server, so expect it to take a few minutes and a few GB of disk. Later starts reuse the built image and come up in seconds.
This starts the Veda server, MinIO blob storage, and the NATS message queue:
| Service | URL |
|---|---|
| Veda API | http://localhost:6340 |
| Niriksha dashboard | http://localhost:5173 |
| MinIO console | http://localhost:9001 |
Optional profiles
Add capabilities with Compose profiles:
# With embedding service (HuggingFace TEI)
docker compose --profile embedding up
# With document OCR (FlashParse + GLM-OCR, text-only, ~8GB VRAM)
docker compose --profile ocr up
# With full OCR + VLM image captioning (~48GB+ VRAM)
docker compose --profile ocr --profile ocr-vlm up
# With self-hosted web crawl + meta-search (crawl4ai + searxng)
docker compose --profile web up| Profile | Services | GPU VRAM | Purpose |
|---|---|---|---|
embedding | HuggingFace TEI + Multimodal Embedding (vLLM) | CPU + ~10GB | Text embedding (BGE-M3) + multimodal embedding (Qwen3-VL) |
ocr | FlashParse + GLM-OCR | ~8GB | Text/table/formula OCR from documents |
ocr-vlm | VLM (Qwen3.5-35B-A3B) | ~48GB | Image captioning and classification (use with ocr) |
web | crawl4ai + searxng | CPU | Self-hosted web crawl + meta-search for web connectors & agentic web search |
monitoring | OTel Collector, Prometheus, Mimir, Tempo, Loki, Grafana | CPU | Metrics, tracing, and log aggregation |
Authenticate
Veda uses JWT authentication with role-based access (Admin, Editor, Viewer). Set VEDA_BOOTSTRAP_ADMIN_PASSWORD before the first start, then obtain a token:
curl -X POST http://localhost:6340/v1/auth/login \
-H "Content-Type: application/json" \
-d '{"username": "admin", "password": "<your-bootstrap-password>"}'Pass the returned JWT on subsequent requests as Authorization: Bearer $TOKEN.
Create a collection
curl -X POST http://localhost:6340/v1/collections \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $TOKEN" \
-d '{
"name": "products",
"fields": [
{"name": "embedding", "type": "vector", "dimension": 768, "metric": "cosine"},
{"name": "category", "type": "string", "filterable": true},
{"name": "price", "type": "float64", "filterable": true},
{"name": "description", "type": "string"}
]
}'Insert records
curl -X POST http://localhost:6340/v1/collections/products/insert \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $TOKEN" \
-d '{
"records": [
{
"embedding": [0.1, 0.2, ...],
"category": "electronics",
"price": 299.99,
"description": "Wireless headphones"
}
]
}'With the
embeddingprofile running you can skip pre-computing vectors entirely — see auto-embedding.
Search
curl -X POST http://localhost:6340/v1/collections/products/search \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $TOKEN" \
-d '{
"query": {"vector": {"field": "embedding", "values": [0.15, 0.22, ...]}},
"limit": 10
}'Next steps
- Model your data in Collections & fields
- Pick the right index in Indexes
- Go beyond dense search in Search & fusion
- Turn PDFs into searchable vectors with the Document pipeline