LiveContext is a self-hosted automation platform where you describe a job in chat and it builds the automation in front of you: a workflow graph you can read, AI agents with scoped access and budgets, and a small web app for the people who use it. This repository is the Community Edition — the whole platform as a single self-hosted service, free to run in production inside an organisation.
The workflow engine pairs a visual builder with an execution engine supporting parallel branches, loops, signals, human-approval steps, and schedule, webhook, chat, form, and datasource triggers. Each agent has its own model, tool access, files, credit budget, and audit trail, with per-agent metrics charting calls, tokens, success rate, and duration. Built-in tables hold data the workflows and agents read, write, and enrich, and interfaces turn a workflow into forms, dashboards, and approval screens that can be shared as standalone apps.
Deployment is Docker Compose, or an npx livecontext wrapper around it that pulls the images and serves the app on port 3000. The backend is a single Java 21 monolith JAR with PostgreSQL, Redis, an S3-compatible object store, and a tools bridge, alongside a Next.js frontend; images are built for linux/amd64 and linux/arm64. Around 700 integrations are seeded offline at first boot and more can be added as OpenAPI specifications. The first account created becomes the administrator, and organisations and workspaces carry role-based access control.
Two heavier features are opt-in and start no container by default. A renderer profile adds a headless Playwright and Chromium sidecar so interface nodes can produce a PNG screenshot or a PDF, and a browser-agent profile adds a Browser Use container with a SearXNG metasearch sidecar for page browsing and web search. Both are enabled by bundled environment files that ship with the repository, so neither can be turned on through the npx wrapper.
Features
- Workflow engine: a visual builder and execution engine with parallel branches, loops, signals, and human-approval steps
- Triggers: schedule, webhook, chat, form, and datasource entry points into a workflow
- Scoped agents: per-agent model, tool access, files, credit budget, and audit trail
- Chat-built automations: a chat agent designs, builds, and runs the workflow from a described job
- Integration catalog: around 700 integrations seeded offline at first boot, extensible with OpenAPI specs
- Interfaces and apps: forms, dashboards, and approval screens served by a workflow and shareable standalone
- Tables: built-in data tables that workflows and agents read and write, with filter, search, and export
- Metrics: calls, tokens, success rate, and duration charted per agent and per tool
- Single-command stack: one monolith JAR plus PostgreSQL, Redis, object storage, a tools bridge, and the frontend
- Optional renderer: a Playwright and Chromium sidecar for screenshots and PDFs
- Optional browser agent: a Browser Use container and SearXNG sidecar for browsing and web search
- Access control: organisations and workspaces with role-based permissions
