FAQ
The reference pages cover the primitives in depth — components, backends, types, deployments, leases. This page collects the shorter questions that come up first, with answers kept short enough to read in place.
Questions are ordered roughly by what gets asked first. If a question isn't here, the answer is usually on one of the linked concept pages.
Do I have to learn all this myself?
No. The platform is built so an LLM can drive it for you. Install the CLI, run ppl mode general, then hand your agent a task starting with ppl docs get flows/quickstart. The agent reads the docs registry, picks the right flow, runs the bounded commands, and reports back. You answer questions, the agent does the setup. See Agent-mode output and the Quickstart for the hand-off prompt.
Do I need to install anything?
No — the App runs in the browser. You only need a local install if you want to write components in Python or C++, in which case you'll install the ppl CLI.
Which languages can I write components in?
Python and C++ are both first-class. Python is the right choice for ML integrations (PyTorch, ONNX, HuggingFace, Ultralytics). C++ is the right choice for low-latency, high-throughput components — the standard C++ ML SDK ships preinstalled (geometry, image, tensor helpers, Triton client, multi-object trackers).
See the Quickstart for both Python and C++ tracks.
Why does Pipelogic use a strict type system?
Streams between components are typed and checked at wire time. Connecting an Image output to a [BoundingBox] input fails immediately, before anything runs.
Concretely this gives you:
- Safety: mismatches are caught at composition time, not at runtime.
- Compatibility: components from different teams plug together without surprises.
- Documentation: the type signature is the spec.
- Reusability: define a domain type once (
{x: Float, y: Float, depth: Float}), every consumer agrees on the shape.
Can I use my own models?
Yes. Pipelogic ships managed serving services a Component picks via depends_on::
- Multi-model services (one instance, many models) — Triton (ONNX, TensorRT, TorchScript, TensorFlow SavedModel), TorchServe (PyTorch MAR), Ollama (LLM / VLM iteration).
- Parametrized services (one model = one instance) — vLLM (high-throughput LLM serving), SGLang (structured generation, VLMs, OpenAI-compatible).
You can upload your own artifact and bind it to a Component vertex, or point a vertex at a hub model id and let the serving service pull it.
See Models for the model preparation workflow.
Can I work with teammates?
Yes. Workspaces support multiple teammates with role-based access. Components, backends, and files are scoped to your workspace; you can transfer ownership between teammates.
Which browsers are supported?
Recent Chrome, Firefox, Safari, and Edge (Chromium-based). Older browsers may work but aren't actively tested.
Is there a free tier?
Yes. €0/month. 2 collaborators, 500 credits/month (250/day cap), 2 GB storage. Shared cloud runtime only.
Can I run backends on my own hardware?
Yes — on plans that entitle on-premise runtime. You register a Linux machine with your workspace, attach it to a runtime, and deploy. Two registration paths cover the common cases: a networked host that registers itself over HTTPS to the platform, and an air-gapped site through an offline fingerprint approval flow (available on plans that entitle the air-gapped runtime).
Once registered, the same backend you've been running in the managed cloud deploys unchanged onto your own hardware.
What's the difference between the CLI and the App?
Both edit the same underlying backend objects. Use the App when you want to assemble existing components into a working backend; use the CLI when you want to write the components themselves, automate releases, or script ops.
For exact command syntax, run ppl <command> --help.
What's a transformation vs a component?
A component is your code, your dependencies, your model — packaged as an image that runs on a runtime.
A transformation is a typed inlined primitive — flatten, lift_unroll, pack_record, filter, etc. It runs in-process between components, with no image to ship and no container to spin up, so it's free to use.
See Transformations.
What languages does the SDK support?
- Python SDK: CPython 3.10.
- C++ SDK: C++23, handled for you by the standard build system.
How do I report a bug?
First, decide whether it's a bug in your backend (your component crashes, your stream stalls — debug it yourself using the deployment view) or a bug in Pipelogic itself (the App misbehaves, the CLI crashes, a managed runtime fails). Only the second is something you file with us.
Discord #bugs for quick chats, support@pipelogic.ai for anything that needs more space. Include what you were doing, the error message, and how to reproduce.
Where this fits
Each answer above links to the concept page that covers it in full. If none of the questions match what you need, start with the Quickstart.