llms.txt
Compare with other sites' llms.txt files →# llms.txt - AI Model Guidance for Runchat
> Runchat is a visual workflow editor for building custom AI workflows. Generate images, video, audio, and 3D models, build custom tools, and publish workflows as shareable apps and artifacts.
## For AI agents and coding tools
There are three ways to work with Runchat programmatically. For **building or editing a workflow** (adding/wiring/running nodes, publishing a tool), the **MCP server** and **CLI** are usually the easiest — they expose the full canvas toolset as native tools and handle auth for you, so you can connect and start on your own:
- **MCP** (any MCP client): add `https://runchat.com/api/mcp` as a Streamable HTTP server. Claude Code: `claude mcp add --transport http runchat https://runchat.com/api/mcp`
- **CLI** (shell / Claude Code / CI): `npx -y @runchat/cli status` → if it exits 3, `npx -y @runchat/cli login` (opens the user's browser to sign in; waits up to 5 min — run it in the background) → `npx -y @runchat/cli guide` → `npx -y @runchat/cli tools`. Just need an image? Skip the guide: `create_runchat` → `create_image_node --runchat_id <id> --model_id runware:400@2 --params.prompt "..."` → `run_nodes --runchat_id <id> --node_ids <node_id>` (returns the image URL). Pass lists as `--node_ids a,b` and object fields as `--params.prompt "..."` — Windows PowerShell breaks inline JSON. Never ask the user to paste an API key into chat.
The **REST API** works too and is the right choice for **running an already-built workflow** from your own app or product (`POST /api/v1/{runchat_id}`), or for canvas edits from a client that can't use MCP (`POST /api/v1/{runchat_id}/canvas`). See the [API](#api) section below for all three.
## Documentation
Full documentation lives at https://docs.runchat.com. AI-friendly indexes:
- **Docs index** (page list with descriptions): https://docs.runchat.com/llms.txt
- **Full documentation** (all pages combined): https://docs.runchat.com/llms-full.txt
For deep questions about specific features, fetch `llms-full.txt`. The summary below is enough for common questions.
## Key Concepts
- **Workflow (Runchat)**: A visual, node-based graph that processes inputs through AI models and code into outputs. Saved per user.
- **Node**: A unit on the canvas. Built-in node types are Agent, Create, Code, Input, Artifact, Note, and Tool.
- **Edge**: A connection between two nodes that passes typed data.
- **Tool**: A published Runchat workflow that can be reused as a single node, or called by an agent.
- **App**: A Runchat workflow published as a shareable interactive web page.
- **Artifact**: A published page (website, blog post, slide deck) generated from a workflow's outputs.
- **Library**: An installed folder of Tools that show up in the node picker and become callable by the chat agent.
- **Credits**: The unit Runchat uses to bill AI model and tool usage.
## Site Structure
- **/** Marketing landing page
- **/pricing** Subscription plans
- **/contact** Contact form
- **/terms** Terms of service
- **/privacy** Privacy policy
- **/examples** Public examples gallery
- **/editor** Visual workflow editor (auth required)
- **/dashboard** User dashboard for workflows, artifacts, scheduled tasks, and team management (auth required)
- **/play/<id>** Published workflows running as interactive apps (auth required)
- **/artifacts/<handle>/<slug>** Public artifact pages (websites, blogs, slides), no auth required to view
- **/api/v1** REST API for executing workflows and manipulating canvases (Bearer token auth)
## Plugins and Integrations
- **Runchat for Rhino** Rhino 8 plugin with chat panel, viewport screenshot capture, GLB import, Python script execution
- **Runchat for Grasshopper** Use published Runchat workflows as Grasshopper components, plus a chat agent that can read and edit the canvas
- **Runchat for Blender** 3D model generation and viewport capture
- **Runchat for Revit** Image generation from views, custom families and tools
- **Runchat for SketchUp** Workflow integration in SketchUp
- **Runchat CLI** (`@runchat/cli`) Build, edit, and run workflows from any shell or script — the same canvas toolset as the MCP, scriptable for coding agents and CI. `npx @runchat/cli guide`, `npx @runchat/cli tools`
- **Linkable accounts** Fal, OpenRouter, Gemini, Groq, Brave, Tavily, Exa, PureMD, Google Sheets (OAuth)
## Built-in Node Types
- **Agent** Run a single agent turn with a text or reasoning model. Supports tools, structured output formats (Markdown / List / Custom JSON Schema), and tool calling.
- **Create** Run generative media models for images, video, audio, and 3D from Fal, OpenRouter, and other providers.
- **Code** Run JavaScript, HTML, Rhino Python, Blender Python, or Revit C# in a sandboxed or plugin-bridged environment.
- **Input** Text, Image, Number, Toggle, Select, File. Each accepts user input or a connected upstream value.
- **Artifact** Create, update, or fetch a published artifact (website, blog, slides).
- **Note** Annotation/comment node for the canvas.
- **Tool** Any published workflow installed from a library, used as a node.
## Plans
- **Starter** (free): 500 monthly credits, public workflows by default
- **Hobby** ($15/mo): private workflows, save history, BYO API keys, custom LoRA training, team features
- **Pro** ($45/mo): Hobby features plus more credits and discounts, private by default
- **Enterprise**: tiered seats, model whitelist, centralized billing
## Common Use Cases
- **Architectural visualization**: Render styled images from Rhino screenshots, train studio-style LoRAs, generate 3D from concept images
- **Design exploration**: Generate variations across materials, styles, and references in parallel
- **Site research**: Pull Streetview imagery, analyze with agents, build reusable analysis tools
- **Content creation**: Generate websites, blog posts, slide decks, and image artifacts from workflows
- **Automation**: Schedule workflows to run hourly/daily/weekly/monthly, trigger via webhooks
- **Custom AEC tooling**: Generate Rhino/Blender/Revit scripts via chat agent, save and reuse
## API
REST API at `https://runchat.com/api/v1`, Bearer token auth. MCP server at
`https://runchat.com/api/mcp` (Streamable HTTP), OAuth 2.1 or Bearer token auth.
- **MCP** — full workflow + canvas tools over the Model Context Protocol
- `POST /api/mcp` — `list_runchats`, `create_runchat`, plus all Canvas tools
- **Execution** — run a workflow as a black box
- `GET /api/v1/{runchat_id}/schema` — fetch input/output schema
- `POST /api/v1/{runchat_id}` — execute with inputs
- `GET /api/v1/{runchat_id}` — execute with no inputs (scheduled / no-input workflows)
- **Canvas** — build/edit the workflow graph itself
- `POST /api/v1/{runchat_id}/canvas` — tool calls like `get_canvas`, `create_prompt_node`, `create_code_node`, `create_image_node`, `connect_nodes`, `run_nodes` (per-type creation tools; the older polymorphic `create_node` still works as a legacy alias)
Use Execution when treating a workflow as a callable tool. Use Canvas when building, editing, or inspecting the graph (scratchpad, moodboard, programmatic authoring).
A machine-readable skill document for the Canvas API (the full tool schema, auto-generated) is at `https://runchat.com/api/skills/canvas`. Authentication is documented at `https://runchat.com/auth.md`.
- **CLI** — `@runchat/cli` runs the same canvas + workflow toolset as the MCP from any shell. `npx @runchat/cli tools` lists every command, `npx @runchat/cli <tool> --help` reads its parameters, `npx @runchat/cli guide` prints the workflow-building guide. Auth: `npx @runchat/cli login` signs in through the browser; `RUNCHAT_API_KEY` for CI. `npx @runchat/cli setup claude` installs a Claude Code skill. Best for coding agents and CI.
## Agent Discovery
Well-known resources for autonomous agents:
- **Authentication** — https://runchat.com/auth.md
- **MCP server card** (SEP-1649) — https://runchat.com/.well-known/mcp/server-card.json
- **Canvas API skill** (machine-readable) — https://runchat.com/api/skills/canvas
- **Skills index** — https://runchat.com/.well-known/agent-skills/index.json
- **API catalog** (RFC 9727) — https://runchat.com/.well-known/api-catalog
- **Protected resource** (RFC 9728) — https://runchat.com/.well-known/oauth-protected-resource
- **Authorization server** (RFC 8414) — https://runchat.com/.well-known/oauth-authorization-server
Pages also support markdown content negotiation — send `Accept: text/markdown`.
## Privacy and Access
- Public marketing pages and artifacts are crawlable
- Workflows are public by default on Starter and Hobby; private by default on Pro
- API endpoints require authentication
- Credentials are encrypted and only sent to user-defined allowed origins
## Audience Notes
- Most users are AEC, design, or creative pros — default to that context when unspecified
- The in-app chat agent can read the canvas, create and connect nodes, edit code, and run workflows
- Workflows are visual graphs; think in nodes, edges, and typed data