Definition

Multica is TeamCity for agent work

Multica offers a practical way to make this transition: it moves agents, pipelines, runs, and artifacts into a company-managed system outside the developer’s laptop. It is not a new model or yet another chat UI, but a platform for managing digital executors and their work.

TeamCity separates a CI job from a developer’s laptop: it provides a server, executors, queues, statuses, and artifacts. Multica does the same for agent runs.

The threshold for a local pilot is low. Multica can be deployed inside the company’s environment; no commercial license is required for internal use by a single organization. This does not mean connecting another external SaaS: code and data remain in the organization’s own environment. Only the models and tools used by the agents themselves need separate controls.
License clarification

The repository does not use unmodified Apache 2.0, but a modified version with additional terms. Hosting Multica as a service for third parties or embedding it in a commercially distributed product requires a commercial license; removing interface branding requires separate written permission. When using only the backend, daemon, or CLI without the interface, attribution notices must be preserved and the product must state that it is built on Multica. This is a summary; the LICENSE text is legally controlling.

At the technical level, the platform consists of a TypeScript frontend, a Go backend, and a Go daemon that runs on runtime machines—the equivalents of workers or runners. Codex, Claude, Hermes, Qwen, and other harnesses used to launch agents are installed on these machines.

Three technical layers
Control plane

Frontend

A TypeScript application for agents, skills, tasks, queues, artifacts, metrics, and observability.

Coordination

Backend

A Go service that stores entities and manages events, integrations, and run state.

Execution plane

Daemon

A Go process on the runtime: it accepts a task, launches the required harness, and collects the execution trace.

The main change is organizational, not technological: an agent run gains an address, identity, state, history, and owner. It stops being a temporary chat on one computer.