Skip to content

Install datasight

datasight is distributed as a Python CLI on PyPI. The recommended installer is uv, which handles Python version management and installs datasight as a global tool. If you already have a Python toolchain you prefer, pip works too.

# Install uv (one-time — see https://docs.astral.sh/uv/getting-started/installation/)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install datasight
uv tool install datasight

pip install --user datasight

Both options include support for DuckDB, SQLite, PostgreSQL, and all AI providers (Anthropic, OpenAI, GitHub Models, Ollama) — no extra packages needed.

Configure an AI provider

datasight needs an AI provider to translate your questions into SQL. If you're choosing for the first time, start with Anthropic — it's the default, the models are good, and you can get a key in about two minutes. GitHub Models is a free alternative if you have a GitHub account.

Run datasight config init to create a shared credentials file that every project on this machine will pick up automatically:

datasight config init
# Creates ~/.config/datasight/.env — edit it to add your key

Alternatively, paste the key directly into a project's .env file instead.

  1. Go to console.anthropic.com → API Keys → Create Key.
  2. Copy the key (starts with sk-ant-).
  3. Add it to your .env:
ANTHROPIC_API_KEY=sk-ant-...

No other settings are needed — anthropic is the default provider and Claude Haiku is the default model, which handles SQL generation well at low cost.

Anthropic accounts require a payment method

New accounts come with trial credit, but you'll need to add a payment method before that runs out. If you'd rather skip billing entirely, see GitHub Models below — it's free with just a GitHub account.

LLM_PROVIDER=openai
OPENAI_API_KEY=sk-...
OPENAI_MODEL=gpt-4o-mini

For Azure OpenAI or a corporate gateway, also set OPENAI_BASE_URL.

GitHub Models provides free access to GPT and other models using your GitHub account — no billing setup required.

LLM_PROVIDER=github
GITHUB_TOKEN=ghp_...
GITHUB_MODELS_MODEL=gpt-4o

To get a GITHUB_TOKEN:

  • GitHub CLI (quickest): run gh auth token and paste the output.
  • Personal access token: go to GitHub → Settings → Developer settings → Personal access tokens → Fine-grained tokens → Generate new token. Grant the Models: read account permission. Classic tokens do not work.

Free tier context limit

GitHub Models caps requests at 8,000 tokens. Databases with more than ~20 tables can exceed this. If you hit "request too large" errors, see Limit schema sent to the LLM.

Ollama runs AI models on your own hardware — nothing leaves your machine and there's no per-query cost. Use it when data sensitivity or offline use requires it; for most users, a hosted provider gives better results with less setup.

Install Ollama, then pull a tool-calling model:

ollama pull qwen2.5:7b

Then configure .env:

LLM_PROVIDER=ollama
OLLAMA_MODEL=qwen2.5:7b

qwen2.5:7b works well for CLI queries (datasight ask) and is the safest cross-platform default — it uses ~2 GB resident on Apple Silicon, so it fits even on 16 GB Macs. For the web UI with chart generation, qwen2.5:14b handles the more complex interactions better. On Apple Silicon with 48 GB+ of unified memory, qwen3.6:35b-a3b-coding-mxfp8 produces noticeably richer answers with comparable decode speed (sparse MoE). See Choosing an AI provider for measured per-model memory footprints and Apple-Silicon-specific guidance.

See the Configuration reference for every supported variable.

PNG chart export

datasight ask --chart-format png needs the optional export extra. Reinstall with uv tool install "datasight[export]" or pip install --user "datasight[export]". The web UI does not need it.