Your app.
Now with an agent.

Drop an AI agent into the product you already ship. Chat, tools, and a sidecar runtime — free, open source, and your routes stay yours.

Get started Host SDKs

prompt Install Sveda into this project. Read https://sveda.dev/llms.txt first.

Install Sveda into this project.

Read:
- https://sveda.dev/llms.txt
- https://sveda.dev/docs/start
- https://sveda.dev/docs/hosts

Pick the matching host SDK from /docs/hosts. If unsure which backend or frontend this repo uses, ask the user.

Apache 2.0 · Laravel · Django · Rails · Next · Spring · or a static page

Vue React Svelte Solid

1 · App.vue

<template>
  <SvedaChat />
</template>

2 · AppServiceProvider Laravel SDK

SvedaClient::host()
  ->resolveToolsUsing(fn () => [
    new FindAccountTool,
  ]);

Your tool. The agent can call it.

YOUR CRM · SALES

Find the Acme account.

FindAccountTool find_account → #10482

Opened Acme Corp. Owner John D. Stage: Negotiation. Last invoice INV-8821 unpaid.

Ask anything…

How it works

Self-host or Cloud. A widget on the page. Tools on your backend. That is the integration.

01

Run Sveda

Self-host :8787

docker compose up

or

Cloud Coming soon
02

Drop it on the page

your.app

sveda-chat
03

Declare a tool

BACKEND tool

+ FindAccount ok
+ GetInvoice ok

your API
04

That is it

LIVE <sveda-chat>

FindAccount

GetInvoice

The agent is in the product.

LIVE ON THIS PAGE

This site already embeds the same agent. It searches the docs — open it and ask.

The stack you already run.

Mint a session from the backend. Open chat on the page. Not a React-only copilot and not a new TypeScript app.

Vue, React, Svelte, Solid.

Four packages, one chat. Import styles.css, pass the embed token, and mount SvedaChat.

Auth stays on the host.

Widget on the page. Session mint on your backend. Agent loop on the sidecar.

PAGE

Page

Your UI in the browser.

your.app

sveda-chat

embed token only — no host key
HOST

Your app

Auth, routes, and session mint stay yours.

YOUR APP auth

POST /sveda/session

your APIs as tools
SIDECAR

Sidecar

Beside the host — not in your request cycle.

SVEDA-SERVER :8787

POST /sveda/stream

agent loop · MCP · catalog
MODELS

Models

Keys on the runtime. Swap providers without a UI rewrite.
BYOK
- ChatGPT responses
- Claude anthropic
Failover keeps the stream going.

Rust runtime →

sveda --tools

IN THE PRODUCT

Tools on the wire.

Builtin search, host MCP, and frontend tools share one stream. No second orchestration repo.

host MCP

Your backend APIs become agent tools.

client tools

The agent clicks, fills, and navigates in the browser.

search_code

Find files in the workspace index.

read_code_file

Open what it found, then answer.

spawn_tasks

Parallel work in one turn.

search_agent_tools

Discover more MCP tools on demand.

THIS TURN

your API mcp
your UI js
workspace optional

Not another chatbot SKU.

Teams want an agent inside the product they ship — without replacing the app or standing up a control plane.

Chatbot SaaS

Lives next to the product, not in it. Your APIs are an afterthought. Auth is someone else’s.

Agent frameworks

You still assemble the UI, the stream, the auth, and the deploy. Primitives — not a product agent.

React-only copilots

Fine on Next. Awkward on Laravel, Rails, or Spring. The rest of the stack does not disappear.

Homegrown glue

Control plane, stream parser, tool loop, model keys. Weeks before a visitor can send a turn.

sveda --delta

SHIP

Roll your own

6-8 weeks

New agent repo, then a widget, then auth
Stream parser and tool loop by hand
Model keys and failover in a third place

Sveda

1 evening

Docker sidecar next to the app
Host SDK mints the session
Drop-in chat that can call your tools

Paste a prompt. Get an agent in the repo.

Sveda is free and open source — Apache 2.0, self-hosted. Or skip the install and ask the agent already on this page. It searches the docs.

Get started GitHub

FAQ

What is Sveda?

Sveda is a free, open-source AI agent you embed in an existing product. Drop in chat UI, connect your APIs as tools, and the app can act — without replacing it or standing up a separate agent platform. Apache 2.0 licensed and self-hosted.

Do I replace my app?

No. sveda-server runs beside the host. Your routes, auth, and deploy stay yours. The page opens chat; the sidecar owns the agent loop.

Which stacks does it support?

Host SDKs cover PHP, Laravel, Python (Flask, Django, FastAPI), Node (Express, Nest, Next), Ruby, Go, Java, and .NET. No framework: static HTML plus a tiny POST /sveda/session. Frontend is Vue, React, Svelte, Solid, a <sveda-chat> web component, or the headless JS client.

What can the agent actually do?

Stream chat in your UI, call frontend tools in the browser, and call your backend over MCP. Optional workspace tools can search and read code. Models and failover live on the sidecar, not in the widget.

Which models can I use?

Put an OpenAI (ChatGPT) or Anthropic (Claude) key on the sidecar, or point a catalog row at any Responses or Anthropic-compatible URL. DeepSeek is one optional wrapper we ship. Failover is an ordered list — swap a provider without rewriting the UI.

Can I install Sveda with Cursor?

Yes. Copy the agent prompt from the homepage or /docs. Each host SDK page has a stack-specific prompt (Laravel + Vue, Next, Rails, Spring, …). Point the agent at https://sveda.dev/llms.txt.

Is this a hosted chatbot?

Open-source Sveda is self-hosted: you run sveda-server beside your app. This marketing site is the brochure. Optional Sveda Cloud, coming soon, is a separate console and hosted runtime. See /cloud.

What is Sveda Cloud?

Coming soon. A console, model tokens on the subscription, and a private runtime per project. Embed the chat in web, iOS, and Android. Teams can add SSO, SCIM, audit export, spend caps, and a key they hold. See /cloud.

Your app.Now with an agent.