Deployment Overview

Tribunus supports multiple deployment models depending on your requirements for data residency, latency, scaling, and infrastructure control. The deployment architecture centers on the Compute Engine, which hosts agent runtimes and model inference backends.

Self-Hosted Compute Engine

The Compute Engine is the server-side component that runs agent sessions, executes tools, and mediates communication with model providers. It is distributed as a single static binary with a SQLite-backed event store.

Hardware requirements vary by workload. For light agent usage with cloud-hosted models (OpenAI, Anthropic), a 2-core, 4 GB RAM instance is sufficient. For local inference with large language models, a GPU instance with at least 16 GB VRAM is recommended.

Deployment involves downloading the binary, setting environment variables for backend configuration, and starting the service. The Compute Engine exposes a REST API for client connections and a WebSocket endpoint for streaming agent output.

Desktop Agent Mode

In Desktop Agent mode, the Tribunus Desktop application runs agents locally without a remote Compute Engine. This is the simplest deployment: install Desktop on macOS, configure your model backends in the settings, and begin working with agents.

Desktop Agent mode stores session data locally. For team deployments, you can configure Desktop to point at a shared Compute Engine instance, centralizing session storage and enabling collaborative agent workflows.

Cloud Integration Patterns

Tribunus can be integrated with existing cloud infrastructure. Common patterns include:

• Sidecar deployment: run the Compute Engine alongside an existing application to provide agent capabilities as a local service
• API gateway integration: expose agent endpoints through an existing API gateway for authentication, rate limiting, and observability
• Event-driven agents: trigger agent sessions from webhooks, message queues, or cron schedules for automated workflows

Production Considerations

When deploying to production, consider the following:

• Authentication: secure the Compute Engine REST API with a bearer token or integrate with your identity provider
• Rate limiting: configure per-user or per-session rate limits to protect model API budgets
• Observability: the Compute Engine emits structured logs and OpenTelemetry traces for integration with monitoring systems
• Backup: the SQLite event store should be backed up regularly; for high availability, configure an external PostgreSQL database