1. Prerequisites

• Python 3.11+
• An Anthropic-compatible gateway URL + API key (the examples use http://35.220.164.252:3888)
• Optional: MySQL or PostgreSQL client libs — only if you point the db capability at one of them.

Tips

DataMind v0.2 does not require the claude CLI or claude-agent-sdk. We talk to the gateway via the official anthropic Python SDK — some environments ship a vendor-rebranded claude binary that ignores ANTHROPIC_API_KEY, so we sidestep that entirely.

2. Install

git clone https://github.com/your-org/DataMind.git
cd DataMind
python -m venv .venv && source .venv/bin/activate

# Install the v0.2 package
pip install -e .

# Optional extras
pip install -e '.[mysql]'        # pymysql + cryptography
pip install -e '.[huggingface]'  # sentence-transformers for local embeddings
pip install -e '.[dev]'          # pytest + pytest-asyncio

3. Configure

cp .env.datamind.example .env.datamind
$EDITOR .env.datamind

Minimum required:

DATAMIND__LLM__API_BASE=http://35.220.164.252:3888
DATAMIND__LLM__API_KEY=sk-your-key-here
DATAMIND__LLM__MODEL=claude-sonnet-4-6

That's it. The same key drives embeddings too — the embedding provider auto-falls-back to LLM credentials if DATAMIND__EMBEDDING__* isn't set.

4. Verify connectivity

python -m datamind.scripts.hello_sdk

Expected:

[hello_sdk] gateway = http://35.220.164.252:3888/
[hello_sdk] model   = claude-sonnet-4-6
[hello_sdk] prompt  = 'Reply with just the single word: pong'
[hello_sdk] --- stream ---
pong
[hello_sdk] usage: input=16 output=5 cache_read=0 cache_create=0
[hello_sdk] OK: gateway reachable, streaming works, model replied 'pong'.

If this fails, the rest will too — fix the API key or base URL first.

5. Try every capability standalone

Each capability ships with a live smoke script. They use a throwaway profile (hello_<cap>_demo) so they don't touch real data.

python -m datamind.scripts.hello_kb       # Chroma + hybrid retriever
python -m datamind.scripts.hello_db       # SQLite + NL2SQL + safeguards
python -m datamind.scripts.hello_graph    # NetworkX multi-hop (no LLM needed)
python -m datamind.scripts.hello_skills   # SKILL.md semantic search
python -m datamind.scripts.hello_memory   # short + long term + fact extraction
python -m datamind.scripts.hello_agent    # the full agent — 4 real questions

6. Run the full agent

# Interactive REPL
python -m datamind chat

# One-shot question
python -m datamind ask "How do I review a pull request?"

# Build / rebuild the KB vector index for the active profile
python -m datamind ingest

# Show config and registered tools
python -m datamind info

7. HTTP server + browser UI

python -m uvicorn datamind.server:app --host 127.0.0.1 --port 8000

Open http://127.0.0.1:8000 in your browser for a chat UI with streamed answers, collapsible tool-call cards, and a sidebar inspector (config, tools, graph stats, KB docs, memory viewer, one-click reindex).

Or talk to the API directly:

Method & path	Purpose
GET /	Browser UI (served from static/app.html)
GET /api/health	Liveness + config snapshot
GET /api/tools	Every registered tool's name, description, and JSON schema
POST /api/ask	Non-streaming convenience
POST /api/chat	Real SSE stream of text / tool_use / tool_result / done events
POST /api/kb/reindex	Rebuild the KB
GET /api/kb/documents	Docs under the active profile
GET /api/memory/{namespace}	Peek at a memory namespace
GET /api/graph/stats	Node / edge counts

Quick check:

curl -s http://127.0.0.1:8000/api/health | jq
curl -s -X POST http://127.0.0.1:8000/api/ask \
  -H 'Content-Type: application/json' \
  -d '{"message":"Say 你好."}' | jq

Stream:

curl -N -X POST http://127.0.0.1:8000/api/chat \
  -H 'Content-Type: application/json' \
  -d '{"message":"Tell me the Status meeting time."}'

7a. (Optional) Swap the agent loop to claude-agent-sdk

DataMind ships two interchangeable agent-loop implementations. Toggle with one env var:

Backend	Path	Pick when
native (default)	Pure Python, anthropic SDK → your gateway	Simplest deploy, fewest deps
sdk	claude-agent-sdk → claude CLI → CCR → your gateway	You want Hooks / Subagents / Compaction / Plan mode

The 23 DataMind tools, SSE event protocol, and frontend are identical on both — only the inner loop differs.

Why CCR

The SDK only speaks Anthropic's /v1/messages. If your upstream is OpenAI-format (/v1/chat/completions), drop claude-code-router (CCR) in the middle — a tiny Node process that translates both directions, ~20ms overhead per request.

Start CCR

# Needs node >= 18
export UPSTREAM_BASE=http://your-gateway.example.com/v1
export UPSTREAM_KEY=sk-...
export UPSTREAM_MODEL=claude-sonnet-4-6

bash scripts/start_ccr.sh
# → listens on http://127.0.0.1:13456

Keep it running in its own terminal.

Switch DataMind to the SDK backend

# In .env.datamind or exported inline:
export DATAMIND__AGENT__BACKEND=sdk
export DATAMIND__AGENT__CCR_BASE_URL=http://127.0.0.1:13456

# Everything else unchanged:
python -m datamind chat
python -m uvicorn datamind.server:app --port 8000

Server startup logs show the backend:

INFO agent_loop_backend backend=sdk ccr=http://127.0.0.1:13456

Switch back to native any time with DATAMIND__AGENT__BACKEND=native (or unset — it's the default).

8. Run the test suite

pytest datamind/tests/
# 95 passed in ~0.6s — no network required

Legacy v0.1 still works

If you want to compare, the old main.py / server.py / modules/ layout remains untouched. It still reads the original .env keys (LLM_API_BASE, LLM_API_KEY, etc.).

Compatible gateways & models

Any Anthropic-compatible /v1/messages service works. Confirmed against http://35.220.164.252:3888 with:

Model	Good for
claude-opus-4-7	Complex reasoning subagents
claude-sonnet-4-6	Main agent (default)
claude-haiku-4-5-20251001	Memory fact extraction, cheap subtasks

Troubleshooting

Symptom	Fix
ValidationError: llm.api_key: Field required	Set DATAMIND__LLM__API_KEY in env or .env.datamind.
401 Invalid token	Key / gateway mismatch; verify with curl $BASE/v1/messages.
Unknown embedding provider 'openai'	Run from the repo root so datamind is importable.
OperationalError: no such table: employees	You hit db_query_nl before seeding; run hello_agent.py or seed manually.
Agent not ready from HTTP server	Lifespan still warming up; wait a few seconds, retry /api/health.