Define the agent in markdown
One file: instructions, model, guardrails. No framework, no boilerplate.
The embedded agent harness for in-product AI
Distri drives real AI experiences inside your product. Describe an agent in markdown, hand it a few of your own functions, and it fills the form, reconciles the ledger, edits the record, inside your UI, streaming every step.
curl -fsSL https://distri.dev/install.sh | shConnect your client-side tools
Expose your product’s own functions. The agent decides when to call them; your code decides what they do.
Let it drive your product
Drop in the chat widget and the agent works inside your UI, streaming every step. Minutes, not quarters.
Showcase
Embedding agents in your product flows
An incident report, filed from a single paragraph of prose. The agent calls the form’s own functions in a few lines of code using @distri/react.
Using Distri
One agent definition. Reach it from the terminal, your server, your UI, or plain HTTP.
# Installcurl -fsSL https://distri.dev/install.sh | sh
# Run an agent against a taskdistri run --task "File a phishing report for John Smith" \ --agent form_filler
# Chat with it interactivelydistri tui form_filler
# Serve it as a streaming API (listens on :7777)distri serve
# Ship the whole agents/ skills/ templates/ tree to your workspacedistri pushdistri push --dry-run # preview what would uploaddistri push skills/my-skill # or push a single item
# Or push one resource at a timedistri agents push agents/form_filler.mddistri skills push skills/ --alldistri prompts push templates/system.hbs
# Pull the workspace back down into the same layoutdistri checkoutA full-featured agent harness
Everything you need to run agents inside a real product, not just a chat box.
In-product tools
Hand the agent a few of your own functions. It pans the map, fills the field, updates the grid, and your UI responds live. You decide what each function does; the agent decides when to call it.
In-product tools →Maps · live
Ready-to-go widgets
Chat, tool cards, task views and sub-agent trees ship as React components. Drop them in as they are, or build your own interface on the SDK, themed to match your product.
React integration →Embedded assistant
Traces
Every run is an OpenTelemetry trace: tool calls, tokens, latency, reasoning. Group, tag and search them, so you can debug and cost your agents in production instead of guessing.
Traces & observability →Traces
Skills
Package reusable capabilities as skills and share them across agents, or pull more in from a registry with a single command.
Agent skills →Any model
Run on GPT-5.4, Claude, or popular open-source models like Qwen. Swap the model in one line of the agent definition.
Model settings →Agent management
Define agents as markdown and push them to your workspace. Manage models, secrets, connections and usage from the CLI or the dashboard.
Managing agents →Products powered by Distri
Real products using in-product agents today.
Embeds agents directly into Google Sheets: analyse data, generate formulas and automate workflows in natural language, with context held across cells and sheets.
Read case study →
An AI copilot for browser automation. Agents drive live browser sessions, store observations in session state, and guide users through building automations.
Read case study →
Teacher agents score student writing with custom evaluation tools, give inline feedback, and adapt to each student, with one thread per student, per lesson.
Read case study →Put an agent inside your product
Define it in markdown, hand it your functions, ship it today.