Trinity

Sovereign infrastructure for stateful AI agents

Self-hosted runtime for stateful AI agents. Reachable from Slack, Telegram, WhatsApp, and Claude Code, with tamper-evident audit on every action.

View on GitHub Request a demo

Independently security-audited by

See Trinity in action

5-minute overview of what Trinity does and why it matters.

Subscribe on YouTube

Platform tour

A visual walkthrough of Trinity's key capabilities.

Multi-agent orchestration

Visualize your entire agent ecosystem in real time. The graph view shows specialized agents, their handoffs, and live status. Tag any agent directly from Slack or Telegram — orchestrator optional.

Diagram showing Trinity at the center connected to Slack, Telegram, WhatsApp, public webhooks, and Claude Code

Multi-channel agent access

Talk to your agents from Slack, Telegram, WhatsApp, or public webhooks. Cross-channel access is governed by a verified-email allow-list, so the same agent can be reached safely from any channel your team uses.

Trinity Schedules tab showing cron-based scheduled tasks for an agent with execution history

Scheduled agent runs

Cron-based schedules let agents work without you in the loop. Set weekly research sweeps, daily campaign builds, or hourly health checks — pause, resume, and inspect every past run from execution history.

Trinity timeline view showing agent collaboration and task handoffs

Timeline & observability

Watch agents collaborate in real time. Track handoffs, filter by event type, zoom into a time range, and replay any execution end-to-end.

Deploy whole agent systems in one command

Trinity ships with scaffolds for complete multi-agent setups — pre-configured systems for sales, marketing, support, and operations. One command to deploy. They run on your Trinity from day one and adapt to your business.

$trinity install <scaffold>

See packaged solutions

Why enterprises choose sovereign agent infrastructure

The security, governance, and control that production AI deployments require. No vendor lock-in. No data leaving your perimeter.

Sovereign

Your infrastructure, your perimeter. Self-hosted on your hardware. Agents persist their state to your GitHub.

Connected

Reach your agents wherever your team works. Slack, Telegram, WhatsApp, public webhooks, Claude Code.

Governed

Tamper-evident audit log. Approval queue for human review. Role-based access. Per-agent guardrails and limits.

Observable

Watch agents work in real time. Replay any execution. Track cost and health per agent.

Learn more about how it works

Diagram showing a Trinity instance and its agents inside a self-hosted perimeter, each agent connected to its own GitHub repository

Sovereign — your infrastructure, your perimeter

Your infrastructure stays yours. Trinity runs inside your perimeter on your hardware. The only outbound dependency is the Git host you choose for agent state — GitHub.com or your self-hosted Git.

Self-hosted Docker runtime with local SQLite — no external database required.
Each agent owns a Git repo it commits to. Per-agent personal access token, branch ownership enforcement, parallel-history detection.
Reset-to-main-preserve-state safe recovery and an auto-sync heartbeat keep agent state durable across restarts.
Self-hosted Git URL support — works with GitHub Enterprise, internal GitLab, and self-hosted Gitea.

Connected — reach agents from any channel

Trinity meets your team where they already work. Slack, Claude Code, WhatsApp — same agents, same audit log, same controls.

Slack

Socket Mode and webhook transport, multi-agent channel routing.

DMs with voice transcription.

Twilio-based DMs (bring your own Twilio account).

Public webhooks

HTTP triggers for any scheduler, secured with endpoint tokens.

Claude Code via MCP

Dozens of tools across modules covering agents, chats, schedules, executions, audit, monitoring.

On identity-bound channels (Slack, Telegram, WhatsApp, Claude Code), a verified-email allow-list governs who can reach which agent. Public webhook endpoints are secured separately with endpoint tokens.

Build locally, deploy remotely

Build and test agents locally with Claude Code, then push to your Trinity instance with Trinity Sync.

Proactive agent messaging

Agents can reach verified-email users directly via the send_message tool — useful for status updates and approvals.

Monetize agents

Public agent links with configurable per-request pricing via Nevermined x402.

Diagram of a tamper-evident audit log as a hash chain of action records ending at a verification endpoint

Governed — production controls for stateful agents

Production-grade controls for stateful agents. You decide what they can do, when humans intervene, and what gets recorded.

Tamper-evident audit log of every agent action: append-only with database-level triggers, hash-chained with a verification endpoint, CSV/JSON export. Coverage includes all MCP tool calls and operator approvals.
Approval queue for human-in-the-loop review on sensitive actions.
Role-based access: admin / creator / operator / user, whitelist-driven defaults.
Per-agent guardrails, execution timeouts, parallel capacity limits, read-only mode, autonomy toggle.