ChatOllama

ChatOllama is an open source agentic app for running AI agents across local and
hosted models. Built on Pi Durable, it provides a SvelteKit web interface
for persistent agent conversations, document knowledge bases, and tools. Data is
stored locally in SQLite and files, with no separate database or vector service
required.

Features

• Chat with your choice of model: connect AI providers or local models through
  OpenAI-compatible endpoints, including Ollama, and switch models within a conversation.
• Persistent conversations: stream responses, inspect tool activity, and keep
  tasks running after closing the browser.
• Document knowledge bases: upload PDF, Markdown, and plain-text documents,
  then ask questions with source citations.
• Agent tools: use built-in file and shell tools, or connect local and remote
  tools through Model Context Protocol (MCP).
• Reusable skills: install skills and choose which ones are available in each
  conversation.

Architecture

SvelteKit serves the web interface and API. Pi Durable runs agents and persists
conversation state; Pi AI connects them to local and hosted models.

flowchart TD
  UI["SvelteKit web interface"] <-->|"HTTP / SSE"| API["SvelteKit server API"]
  subgraph App["Node.js application"]
    API --> Runtime["Pi Durable agent runtime"]
    Runtime --> Models["Pi AI provider adapters"]
    Runtime --> Tools["File and shell tools"]
    Runtime --> MCP["MCP integration"]
    Runtime --> Knowledge["Knowledge retrieval"]
    Skills["Installed skills"] --> Runtime
    Runtime <--> Sessions["SQLite conversation storage"]
    API --> Data["SQLite settings and local files"]
    Knowledge <--> Data
  end
  Models <--> Providers["Local and hosted models"]
  MCP <--> Servers["Local and remote MCP servers"]
  Knowledge --> Embeddings["Embedding endpoint"]

Build and run locally

Install Node.js 24 and pnpm 10.8.1, then clone the repository and build
from rewrite/, the SvelteKit application directory:

git clone https://github.com/sugarforever/chat-ollama.git
cd chat-ollama/rewrite
pnpm install
pnpm build

Configure a chat endpoint and start the app. This example uses a running local
Ollama server; replace your-chat-model with an installed model ID:

export HOST='127.0.0.1'
export ORIGIN='http://localhost:3000'
export CHAT_BASE_URL='http://127.0.0.1:11434/v1'
export CHAT_MODEL_ID='your-chat-model'
# Export CHAT_API_KEY if your provider requires an API key.

pnpm start

Open http://localhost:3000. You can add providers and
change models in Settings. Configure a separate embedding endpoint and
model there before indexing knowledge documents.

pnpm start does not load .env automatically; export variables in your shell.
Data is saved in rewrite/data by default. Set DATA_DIR to use another location.
The app has no login, so the example binds it to your local computer.

See the application guide for detailed configuration and
Docker instructions, or the provider guide
for provider authentication.