RailCall
Sign in

RailCall Docs

The local-first runtime for autonomous agents. Start with your station, choose the reasoning you authorize, and review consequential actions before execution.

Quickstart

Install the station on macOS or Linux. The installer drops the railcall CLI, Studio, and the MCP server under ~/.railcall/.

shell
curl -fsSL https://railcall.ai/install.sh | bash

# check the installed release
railcall version

# open Studio; keep this terminal running
railcall studio

Upgrade in place any time with railcall update. Confirm the current release with railcall version (this page is written against station v1.5.24 and Chat 0.12.6, reviewed 9 September 2026).

The station is the runtime; Studio, Chat, and MCP clients are ways to use it. Local protocol use is free. RailHub is the optional hosted layer. Remote model calls, collaborator requests, and integrations depend on the routes and permissions you configure. Read the privacy guide before connecting private data.

For a local receipt demo, run railcall demo in a second terminal. Inspect its output rather than treating a timing estimate as a promise.

Sections

Start here

Chat Guide

Set up Chat, generate a PDF, review an action, and run a supported workflow.

Reasoning

Models & Auto

Why Auto is recommended, what it selects, and how local, shared, and cloud routes differ.

Control

Privacy & Data Boundaries

Model context, action approvals, hosted processing, telemetry, and receipts.

Help

Troubleshooting

Diagnose station connection, model, export, and workflow problems.

Collaboration

Teams & Shared Compute

Connect stations, grant access to a local model, and keep execution governed on the holder.

Developer

Marketplace API

Authentication and selected marketplace endpoints, with request and response examples.

Reference

CLI Reference

Common railcall subcommand, grouped by category — build/audit/verify, market, license, connect, mcp, studio.

Guide

Studio Guide

A guide to the local browser app at 127.0.0.1:8799 — Builder, Sends, Receipts, Modules, Licenses, and the rest.

Integration

MCP Integration

Wire Claude Desktop, Cursor, Windsurf, or Zed to the local RailCall MCP server, and see the tools an AI agent gets access to.

Developer

Marketplace Developer

Publish signed workflows and modules to the RailCall marketplace, and design AI agents that call them safely through MCP.

Source

Install Script + CLI

The install.sh installer and railcall_cli.py live on GitHub. Every SHA in the installer is pinned; drift is refused.