One fast, native app for all four databases you actually use — Azure SQL, SQL Server, PostgreSQL and SQLite — and it's free. Its AI assistant writes the query, charts the results, and explains them in plain English.
brew trust pounceapps/tap
brew install --cask pounceapps/tap/pouncesqlHomebrew 6.0+ asks you to trust a third-party tap once — brew trust does that. Then start the AI with free Google AI, a local model, or a few dollars of OpenAI/Anthropic credits.
Ships with a demo database — open it and start querying, no server or login.
Sign in with your directory — browse subscriptions, servers and databases. No passwords stored.
2017–2025+, local or domain-joined. SQL login, Windows/NTLM, or passwordless Kerberos SSO.
Over LAN, VPN or Tailscale. Pin a whole server and browse every database it hosts.
Open any .db file — no server, no login. Pick the file and query.
Whether you're learning SQL, watching your budget, or wiring up an AI toolchain — it meets you where you are.
Describe what you want and watch the AI build the query — then read it, tweak it, run it. Or write your own and ask "explain this." A patient tutor working on real data.
Free with Google AI. $0 and fully private with a local coder model in Ollama — no data leaves your Mac. Or drop $10–20 of OpenAI / Anthropic credits for the top models. The app itself is always free.
Pair it with PouncePad and Claude Code. Claude drives PounceSQL over the built-in MCP server — querying, diagramming, dropping results into PouncePad — while you watch it happen. One workflow, genuinely powerful.
It runs read-only queries against your live connection, then renders the answer the way you'd actually want it. Read-only by default; writes only when you say so.

Ask in plain language. PounceSQL generates engine-correct SQL, runs it, and turns the result into a Chart.js visual — bar, line, scatter — right in the chat. Every answer shows the exact SQL, with Copy and Use in editor.

"Give me the KPIs" returns clean metric cards and a plain-English summary you can actually paste into a report. Diagrams (Mermaid), heatmaps, Markdown tables and LaTeX all render inline — and export as PNG.
Pin a server and expand it like a tree — every database, schema, table and column, with type badges. Filter across databases; right-click to preview, count, or generate a query.
Syntax highlighting, table- and column-aware autocomplete, and an optional live syntax checker, tuned to the connected engine (T-SQL, Postgres, SQLite). Run the whole editor or just the selection.
Markdown, highlighted code with Copy / Use-in-editor, Mermaid diagrams, Chart.js charts, heatmaps, KPI cards, sortable grids with CSV / JSON export, and LaTeX.
Sync your AI keys, theme and fonts through your own 1Password vault. Push from one Mac, then pull on the next and tick exactly which items to apply — secrets stay encrypted in 1Password.
The assistant and MCP tools never run a write (INSERT/UPDATE/DELETE/DDL) unless you explicitly turn it on. Query with confidence.
Thirteen editor themes, separate fonts for interface and editor, independent sizing for editor, tree, results and status bar, and a collapsible sidebar.

PounceSQL runs a local Model Context Protocol server. One command wires it into Claude Code — then agents can list databases, browse schema, run read-only queries, even take screenshots. You watch every tool call happen live in the Tools panel.
claude mcp add --transport sse pouncesql http://127.0.0.1:7457/sse
From VS Code Dark+ and GitHub Dark to Dracula, Nord, Tokyo Night, and the phosphor-green Retro Terminal — each with a live SQL preview. Pick separate fonts and sizes for the interface, editor, tree, and results.

Connect Settings → Sync to a 1Password vault you choose, then Push your AI keys, theme and fonts. On another Mac, Pull shows a side-by-side list and you tick what to apply — nothing changes until you confirm, and secrets stay encrypted in 1Password.
Pick a provider per your policy — cloud, enterprise gateway, or fully local — and switch models from a dropdown in the chat header. Each provider remembers its own endpoint, model and key.
Direct API key with native tool-calling — GPT, Claude and Gemini can inspect your schema and run read-only queries. Pick any model from the header.
Point at http://host:11434 (localhost, LAN or Tailscale). Your data never leaves your machine — ideal for sensitive work. Load the model list from the server.
OAuth2 gateways with discovery. Routes each model by its native format, shows cost per model, and a PHI-cleared badge — with per-message token counts and session spend.
| Option | Cost | Good for | |
|---|---|---|---|
| Google AI (Gemini) | Free tier | The easiest free start — grab a key in a minute. | Get a key ↗ |
| Ollama (local) | Free · private | Runs on your Mac. Try a coder model — nothing leaves your machine. | Download ↗ |
| OpenAI | ~$10–20 | GPT models. Pay-as-you-go credits stretch a long way. | Get a key ↗ |
| Anthropic (Claude) | ~$10–20 | Claude — excellent at SQL and clear explanations. | Get a key ↗ |
Paste the key in Settings → AI Provider, pick your model, and you're chatting with your data. Switch providers anytime.
Free, native, and notarized for macOS 13+ (Apple Silicon).
brew trust pounceapps/tap
brew install --cask pounceapps/tap/pouncesqlRecommended — stays current with brew upgrade. Homebrew 6.0+ needs the one-time brew trust for third-party taps.
Good to know: the AI assistant and MCP tools are read-only by default. AI can be
wrong — review generated SQL before running, and try a non-production database first. Provided as is.
Azure SQL uses your az login session — see the Azure sign-in help.