Deploy AI agents. Not infrastructure.
Trinity Cloud takes your agent from code to production in minutes. Push your system, get a live runtime with persistent memory, channels, and background tasks — all managed for you.
Current launch surface
This is the frozen early-partner contract Trinity is willing to stand behind today. Anything outside this surface is either selective, guided, or intentionally not part of the launch promise yet.
Standard is the default runtime lane.
Official starter system and manifest-backed import are the supported launch paths.
Releases, per-tenant runtime instances, secrets, tasks, logs, and attachments are part of the core operator loop.
HTTP and Telegram are the supported channel story today.
Firecracker remains the higher-isolation lane, but it is not the default onboarding path yet.
Hosted Slack work exists on trunk, but it is not part of the frozen default operator loop.
Self-serve signup, workspace accounts, org administration, and billing.
Multi-region execution and generalized cloud-platform claims.
A broad channel marketplace or generic app-hosting story.
Who Trinity Is For Right Now
Trinity is in a curated early-partner phase. The product contract is explicit: one supported default lane, two supported launch paths, and clear boundaries on what is not part of the first cohort yet.
Technical teams evaluating agent-runtime infrastructure, not generic app hosting.
Operator agents, workflow agents, and domain-specific assistants that need durable state, tools, and channels.
Teams willing to work inside a guided early-partner model instead of expecting polished self-serve SaaS immediately.
Default lane: standard.
Supported launch paths: official Trinity starter system or manifest-backed import.
Default operator loop: release-backed runtime instances, HTTP + Telegram, secrets, tasks, logs, and attachments.
Teams that need Firecracker as the default onboarding path today.
Teams that primarily need billing, org administration, or self-serve workspace management.
Teams expecting multi-region fleet guarantees or a broad marketplace story on day one.
Three commands to production
Inspect your agent manifest, create a release, and boot a live runtime. No Dockerfiles, no Terraform, no YAML pipelines.
See the starter agent system# Inspect the canonical agent system
$ go run ./cli agent-systems inspect \
--manifest agent-system-sample/agent-system.json
system:
name: agent-system-sample
assistant:
name: trinity-agent
kernel:
kind: agent-core
entrypoint: agent-system-sample/server.mjs
channels:
- http
- telegram
attachments:
- /state durable
- /workspace durable
- /artifacts durable
- /cache ephemeral
# One complete runtime unit:
# kernel + channels + tools + memory + attachmentsEverything your agent needs to run in production
Stop stitching together hosting, state management, and messaging integrations. Trinity handles the platform so you can focus on the agent.
Ship once, run anywhere
Package your agent into an immutable release. Deploy the same artifact across tenants, environments, and regions without rebuilding.
State that survives restarts
Memory, transcripts, and workspace files persist across deploys. Your agent picks up where it left off — every time.
HTTP and Telegram now
The current frozen operator loop supports HTTP ingress and Telegram as the default channel attachment story. Higher-level channel paths can keep maturing behind that baseline.
Secrets and isolation by default
Every tenant gets its own runtime, its own secrets, its own network boundary. API keys never appear in logs or leave the sandbox.
Production-grade security, zero config
AI agents are powerful. Trinity makes sure they stay under control.
Sandboxed runtimes
Every agent runs in its own isolated environment. No shared state between tenants, no cross-contamination between releases.
Locked-down networking
Outbound access is blocked by default. You choose exactly which domains your agent can reach — nothing more.
Secrets that stay secret
API keys are injected at runtime and scoped to each tenant. They never touch disk, never appear in logs, never leave the sandbox.
Full audit trail
Every tool call, every network request, every model interaction is logged and queryable. Know exactly what your agent did and why.
Ready to ship your first agent?
Trinity is in an operator-managed early-partner phase. Launch the official starter system in minutes, then replace it with your own manifest-backed agent system when you are ready.