Raga AI
Canvas, screen by screen
Every page of the Canvas app: projects, the agent builder and its chat, tools, callbacks, prompts, knowledge bases, tracing, settings, and a workflow builder that hasn't shipped yet.
I built nearly all of it. A few parts came from teammates, and the case study says which.
Captured from the running app with every API response replaced by made-up data for a fictional store. No real customers, people or keys appear.
4 of 61 screens are open. The other 57 are shared on request: send me a short email and I'll reply with a private link.
Projects and sign-in 7
Projects hold a team's agents, tools, prompts and keys. People join one through an invite link.
Agents 3
Every agent in a project, with pinned agents, folders and drag-and-drop to organise them.
Agent builder 9
One screen to configure an agent and test it, with the chat beside the form.
Testing in chat 3
The full-page chat, with saved conversations, tool calls and markdown replies.
Tools 7
Python functions, GraphQL queries and MCP servers an agent can call. Every tool is versioned.
Callbacks 3
Python hooks that run before or after the model, a tool or the agent, such as redacting card numbers or capping refunds.
Prompts 4
Reusable, versioned instructions an agent can import.
Knowledge bases 4
Documents an agent can search, with a query box for checking what it will find.
Tracing to Catalyst 3
Tracing configs send every run to a project in Catalyst, Raga AI's evaluation platform.
Workflow builder (not released) 4
A canvas for multi-step logic with nine node types. It's built, but the sidebar still marks it as coming soon.
Settings and team 6
Profile, collaborators with five roles, model API keys, access keys and project security.
Help 2
Docs and support pages.
Dark theme 6
Some of the screens above in the dark theme.
Want the story behind these screens?



