Evaluate Semurg in an afternoon. Everything here runs on your own machines.

Docker. The image comes from Semurg's own read-only registry at one.semurg.io. The token is printed in the log at first start and kept on the node.

docker run -d --name semurg -p 127.0.0.1:4000:4000 \
  --ulimit memlock=-1 --ulimit nofile=1048576:1048576 --security-opt seccomp=unconfined \
  -v semurg-data:/var/lib/semurg one.semurg.io/semurg
until docker logs semurg 2>&1 | grep -q 'Your API token'; do sleep 1; done
TOKEN=$(docker exec semurg cat /var/lib/semurg/.api_token)

--security-opt seccomp=unconfined is required: Docker's default profile blocks the system calls the engine reads and writes with, and without it the container stops at start with a message naming the flag. --ulimit memlock=-1 lets it register its I/O buffers. Binding to 127.0.0.1 keeps the node on the host until you decide otherwise.

Or install on the machine (a systemd service; the data directory must be on ext4 or xfs):

curl -fsSL https://one.semurg.io/install | sudo bash
TOKEN=$(sudo grep '^SEMURG_API_TOKEN=' /etc/semurg/semurg.env | cut -d= -f2-)

Check it. Without the token the API answers 401; with it, 200.

curl -s http://127.0.0.1:4000/api/health
curl -si -H "Authorization: Bearer $TOKEN" http://127.0.0.1:4000/v1/whoami | head -1

Claude Code

export SEMURG_API_TOKEN="$TOKEN"
claude mcp add-json --scope user semurg '{"type":"http","url":"http://127.0.0.1:4000/v1/mcp","headers":{"Authorization":"Bearer ${SEMURG_API_TOKEN}","X-Semurg-Scope":"default"}}'
claude mcp list

Cursor, Codex and Gemini CLI connect to the same URL; the agents page has each one's configuration.

Python

pip install semurg

The five-line remember / recall / context example is on the PyPI page. Eleven agent frameworks run end to end over MCP, each with a complete program on the agents page: LangChain (Python and Elixir), CrewAI, OpenAI Agents SDK, Microsoft Agent Framework, LlamaIndex, Mastra, Google ADK, Pydantic AI, Vercel AI SDK and Jido.

Text and documents, kept whole and searchable:

curl -s -H "Authorization: Bearer $TOKEN" -H 'content-type: application/json' \
  -d '{"passages":[{"text":"Acme Ltd acquired Widget Co in 2024."}]}' \
  http://127.0.0.1:4000/v1/ingest

Agent memory with a knowledge graph: send text through the MCP memory_remember tool, with the entities and relations your agent found in it as facts; memory_context then answers a question from the graph and cites each fact.

Records with a vector, attributes, text and links (graph edges), queried by similarity plus graph hops:

curl -s -H "Authorization: Bearer $TOKEN" -H 'content-type: application/json' \
  -d '{"op":"hybrid_ingest","d":4,"records":[
        {"id":1001,"vector":[0.1,0.2,0.3,0.4],"text":"graph databases","links":[{"dst":1002,"rel":1}]},
        {"id":1002,"vector":[0.1,0.2,0.3,0.5],"text":"vector search"}]}' \
  http://127.0.0.1:4000/v1/query
curl -s -H "Authorization: Bearer $TOKEN" -H 'content-type: application/json' \
  -d '{"op":"hybrid_query","vector":[0.1,0.2,0.3,0.4],"k":2,"hops":1}' \
  http://127.0.0.1:4000/v1/query

A bulk loader for plain edge lists (millions of edges from a CSV in one call) is in build; until it ships, load edges as record links as above, or ask us and we will load a sample of your data with you.

Graph, in memory (SNAP com-Orkut, 117 million relationships, every engine):

curl -fsSL https://one.semurg.io/bench | sudo bash -s -- --graph

Vector (SIFT1M):

curl -fsSL https://one.semurg.io/bench | sudo bash -s -- --vector

Graph, bigger than memory (SNAP com-Friendster, 1.8 billion relationships, every engine held to the same 12 GB of memory; it needs about 300 GB free on the data disk and most of a day):

curl -fsSL https://one.semurg.io/bench | sudo env BENCH_BAND=exceeds_cap ARENA_MEM_CAP_GB=12 \
  BENCH_DATASET_GRAPH=friendster_cap bash -s -- --graph

The runner checks the machine first and refuses, saying what to do, if a requirement is missing; it never formats or mounts a disk. Its source and method are in the runner itself.