AI & Agents
Treeline is built for AI agents. Your financial data lives in a DuckDB database you own — not a vendor’s cloud — and agents query it directly via SQL: no pre-built API to outgrow, no third party in the middle. Run your agent on the same machine, or let any AI agent that supports remote MCP servers — like ChatGPT or Claude — connect to your data over the network.
Why Local-First Matters for Agents
Section titled “Why Local-First Matters for Agents”Most finance apps store your data on their servers. An agent has to go through a rate-limited API, wait for network round-trips, and work within whatever endpoints the app exposes.
Treeline is different. Your data is a file you control. An agent gets:
- Full SQL access — any query, any join, any aggregation. Not limited to pre-built API endpoints.
- Zero latency — DuckDB runs in-process. Queries return in milliseconds.
- No rate limits — run as many queries as needed for complex analysis.
- Your data, not a vendor’s — it’s a file you hold and can query, export, or move. Agents work with it; they don’t hold it hostage.
This matters more as AI moves into finance. You can wire a cloud assistant to your bank through an aggregator and ask it anything — but the structured data still lives with the aggregator, not you. You’re leasing access: the moment you want the underlying records, or to ask something its history window doesn’t keep (“how has my net worth moved over the last three years?”), you hit the limits of what it retains. Treeline flips the ownership — your data is a file you hold, full history, structured and queryable, and any agent you trust works against it (Claude, ChatGPT, a local model, your own script).
Be clear-eyed about privacy: point a cloud LLM at your data and the results you ask about go to that LLM — the same as the cloud route. The durable advantage isn’t privacy, it’s ownership. Privacy only becomes a real edge when you keep things local — a local model, or no LLM at all — and Treeline gives you that option; the cloud route never does.
How Agents Connect
Section titled “How Agents Connect”Agents reach Treeline in two ways: on the same machine as your data, or over the network through a hub you run.
On your machine
Section titled “On your machine”The agent runs where your data is and connects locally — no hub needed.
MCP Server
Section titled “MCP Server”The Model Context Protocol lets agents discover and use Treeline’s tools over STDIO. Works with Claude Desktop, Claude Code, Gemini CLI, Cursor, and any MCP-compatible client.
The tl command-line interface gives agents direct access to all Treeline features. Every command supports --json for structured output. Best for OpenClaw, Claude Code, and agent frameworks that can execute shell commands.
OpenClaw
Section titled “OpenClaw”The Treeline skill on ClawHub lets you chat with your finances through Discord, WhatsApp, Telegram, and more.
Over the network (via a hub)
Section titled “Over the network (via a hub)”Agents that can’t spawn a local process — ChatGPT, Claude on your phone, claude.ai — reach Treeline through a hub you run, at its /mcp endpoint. Same tools as the local MCP server, authorized with the scopes you approve.
User Skills
Section titled “User Skills”Agents can create and discover skills — reusable financial playbooks stored as SKILL.md files on your machine. A skill might define your tagging conventions, tax categories, or budget targets.
Skills follow the open Agent Skills standard and work across Claude Code, OpenClaw, VS Code, and other platforms that support the spec.