Distri is a composable AI agent runtime for building in-product agents. You define an agent once in a markdown file, connect your product’s own functions as tools, and run it from the CLI, as a streaming API server, or embedded in your app. Unlike a chatbot, a Distri agent reaches into your product and calls your functions — filling a form, reconciling a ledger, updating a grid.

How do I create a Distri agent?

Install the CLI with `curl -fsSL https://distri.dev/install.sh | sh`, write a markdown file with a TOML frontmatter block (name, [tools], [model_settings]) and a system prompt, then run it with `distri run --task "..." --agent NAME`. Local runs need no account — just a model provider key.

How do I deploy a Distri agent?

Authenticate with `distri login`, then sync your agents to Distri Cloud with `distri agents push agents/ --all`. Skills, prompts, and workflows push the same way (`distri skills push`, `distri workflows push`).

How do I embed a Distri agent in my product?

Install @distri/core and @distri/react, wrap your app in DistriProvider, and render a Chat component. Pass your product functions as externalTools so the agent can drive your UI. You can also call the runtime directly with agent.invoke() and client.llm().

Distri is fair-code licensed and always free to self-host. Distri Cloud offers managed hosting with short-lived token auth and usage tracking.

Blog

Building in-product agents with Distri, walkthroughs, samples, and ideas.

Samples

Consolidating three subsidiaries into one statement

Consolidation is where errors hide. Pull three subsidiaries' P&Ls into one set of figures and the numbers will usually look fine, the failure mode is a line that's quietly understated, not one that's obviously wrong. The value of an agent here is that it holds the whole picture and notices the line that doesn't fit the pattern.

Vivek Gudapuri··3 min read

Samples

Clearing a bank statement: what to automate, what to flag

The hard judgement in bank reconciliation isn't matching, it's knowing what to leave alone. A one-cent difference is card-processor rounding; a fifty-dollar difference is a question for a human. A good agent draws that line consistently and shows you only what deserves attention.

Vivek Gudapuri··2 min read

Samples

An agent that reconciles two ledgers, and shows its work

Reconciliation is the kind of work agents are actually good at: mechanical, high-volume, and easy to get subtly wrong. The trick is not just doing the match, it's making the result legible, so a human can trust it in ten seconds.

Vivek Gudapuri··2 min read

Samples

Watch an agent fill a real form

Most "AI form" demos are a video. This one is the real thing: the product component below is the same React form you'd ship, the panel on the right is Distri's real @distri/react transcript, and the run between them is recorded, it plays back through the actual components, so what you see is exactly what a user would see.

Vivek Gudapuri··2 min read

AI-first

Building an AI-first project

The hard part of building an AI product is no longer the model. Anyone can call an LLM in three lines. The hard part is everything around it: the tools the model can call, the knowledge it loads, the parts of its work you can see, and the moments where a human stays in control. That surrounding system is the harness, and in an AI-first product the harness is the product.

Vivek Gudapuri··5 min read