The same C++ runtime meta-harness you built on locally, running on managed infrastructure. Same pipeline definition, no rewrite, no servers to plan for.
Free to start · Subscription plans and metered billing · Self-host anytime
01 · Deploy
One click from the extension, or one command from the terminal — whichever you already live in. Same C++ runtime, same pipeline definition, now on managed infrastructure. Nothing gets rewritten for production, so nothing behaves differently in production.
02 · Compute
The cloud-exclusive model server performs intelligent model loading and sharing across GPUs, request batching, and hot/warm/cold tiering that keeps your busiest models ready.
03 · Observe
Traces, latency, errors and token consumption for every agent, end to end — per member, per team, per org. Replay any run: pause, rewind, watch it as if it were live. When someone asks what the AI spend bought, you have an answer.
04 · Lifecycle
Versioned history and fingerprinted artifacts, so you know exactly which version is in which environment. Compare versions side by side against real data. Rollback is a button, not a rebuild.
05 · Team
Shared frameworks that don't restrict creativity or choice, per-seat access, org-wide spend visibility. The pipeline you built alone becomes the thing your team builds on.
See Team plans →Already building on the open source?
Because your pipeline is a declarative artifact rather than a codebase — nodes and wiring, not glue code — there’s nothing to port. Cloud runs the same C++ runtime meta-harness on the same definition. You’re not migrating an application; you’re choosing where it executes.
And if you change your mind, the exit is the same door as the entrance: pull the pipeline back down and self-host it. No export process, because there was never anything proprietary holding it.
Under the hood
Three things run your pipeline. A routing layer terminates the connection and dispatches work. An execution layer spawns every pipeline as its own OS subprocess — one crashing pipeline cannot take down another, or the service. A GPU-backed model server handles the AI data-intensive workloads: SLMs, OCR, vision, speech, embeddings.
Where the efficiency comes from. The model server is Cloud-only, and it’s where the economics live. Models load once and are shared across GPUs rather than duplicated per job. Requests batch, eliminating the load and unload churn between calls. Hot, warm and cold tiering keeps your busiest models resident and demotes idle ones automatically. GPU capacity scales on actual queue depth rather than on a number you guessed in advance.
You pay for compute that actually executes. Idle pipelines cost nothing.
IDE extension
Deploy from VS Code, Cursor, Windsurf or Antigravity.* One click, no context switch.
CLI
Same deploy, one command, scriptable into whatever CI you already run.
SDK
Call a deployed pipeline from your own app — Python or TypeScript.
MCP
Expose a pipeline as a tool to any MCP client or agent.
* Compatible with any VS Code–based IDE.
A pipeline can sit deployed and idle, waking on a call from your app, your workflow, or an agent — you pay for the run, not the waiting. Or keep it warm when latency matters more than cost. Either way it’s the same pipeline; the difference is a setting, not a rebuild.
Subscription plans and metered billing, with no hard caps, no surprise bills, and no vendor lock-in. Pay for what you run and nothing you don’t.
The billing unit. A token is a fixed slice of the CPU, GPU and memory your pipelines actually consume — not an LLM token. Every plan includes monthly tokens; beyond that, metered billing at $0.03 per token.
The meta-harness underneath Cloud wasn’t written for AI pipelines. It’s a C++ runtime built to move petabytes of enterprise data — process isolation, concurrency and data integrity designed in from the start, not bolted on once traffic arrived. AI workloads inherited that.
Develop on your laptop with the open-source runtime. Deploy the same pipeline to managed infrastructure from your IDE or your CLI — and pay only for what actually runs.
Free to start · Subscription plans and metered billing · Self-host anytime
Rolling this out to a team? Talk to us →
