Chat-to-automation that puts every AI agent on a budget

What it does

LiveContext CE bundles a workflow engine, AI agents, built-in data tables, and a small-app builder into one self-hosted platform. You describe a job in chat and it generates a readable workflow graph, optionally wrapped in forms or dashboards, then dispatches scoped agents to execute it. The whole backend is a single monolithic JAR running under Java 21, paired with a Next.js frontend and the usual Docker suspects: PostgreSQL, Redis, and an S3-compatible store.

The interesting bit

The unusual part is the tight scoping: the workflow itself decides exactly what each agent sees and which tools it gets, plus every agent carries its own credit budget and audit trail. The README claims this cuts token costs compared with “do-everything” agents, and the 700-plus integrations ship fully offline so the platform does not phone home to a vendor catalog on first boot.

Key highlights

• Chat-driven builder: describe the automation in chat and the platform generates the workflow graph.
• Scoped, budgeted agents: per-agent model selection, tool access, credit limits, and full audit trails.
• Built-in tables and interfaces: no external database required for basic data storage; workflows can serve forms, dashboards, and approval screens as standalone apps.
• Monolith architecture: the entire backend ships as one JAR, not a mesh of microservices.
• Heavy features are opt-in: screenshot/PDF rendering and browser-based web search live in separate sidecar containers that add roughly 1–2 GB each and must be enabled via repository-level environment files.

Caveats

• The screenshot/PDF renderer and browser-agent profiles are opt-in sidecars that add roughly 1–2 GB each; the README notes these cannot be enabled through the quick-start wrapper and require a repository clone to supply their environment files.
• An “always-current integration catalog” and hosted account management are exclusive to the managed livecontext.ai service and absent from the CE build.

Verdict

Worth evaluating if you want a self-hosted automation stack where AI agents are scoped and budgeted by default, not bolted on. Look elsewhere if you are unwilling to run Docker Compose or need the managed-only features like hosted account management.