Newcomer Guide
If you are new to AGILab, optimize for one outcome only: one successful local
run of the built-in flight_project from the web UI, including the core
PROJECT -> ORCHESTRATE -> PIPELINE -> ANALYSIS route.
This page gives the mental model only. Quick-Start owns the exact commands. Newcomer First-Failure Recovery owns the first-failure path.
Fast adoption ladder
Use this order when you need the quickest route to confidence:
Stage |
What to do |
Why it matters |
|---|---|---|
Browser preview |
Open AGILAB Demo. |
Confirms the public UI shape before you install anything. |
Local first proof |
Follow Quick-Start with the built-in |
Exercises the real source-checkout install, run, and analysis path. |
Evidence record |
Keep |
Gives support, contributors, and future runs the same baseline. |
Expansion |
Move to notebooks, package mode, private apps, or cluster work. |
Prevents day-1 failures from mixing product, app, and infrastructure variables. |
Choose one route
Goal |
Route |
Use when |
|---|---|---|
See the UI now |
You want a browser-only look at the AGILAB web UI before installing anything. |
|
Prove it locally |
You want the real source-checkout path with |
|
Use the API/notebook |
You want the smaller |
The first proof is deliberately narrow:
use a source checkout, run the built-in flight_project locally from the
web UI, inspect the pipeline recipe, and confirm a visible result under
~/log/execute/flight/.
The landing page first-proof wizard now enforces that same single actionable
route, reads run_manifest.json, and shows a recovery checklist with exact
evidence commands before you branch out.
That is enough for day 1. Do not widen the problem to notebooks, package mode, private apps, or cluster setup until this path works once and the manifest gives you a passing baseline.
This also means PyCharm is not part of the day-1 contract. AGILAB keeps PyCharm run configurations for developers who want IDE debugging, but the newcomer route is shell + browser first. The same install, execute, and analysis path can be driven from commands, the web UI, or checked-in wrappers.
Adoption evidence
On April 24, 2026, the source-checkout first-proof smoke passed locally in
5.86s against the 600s target. On April 25, 2026, the same
source-checkout proof passed on a fresh external machine in 26.87s.
On April 27, 2026, the packaged first-proof CLI passed locally in 7.04s:
uv --preview-features extra-build-dependencies run agilab first-proof --json
The JSON proof writes ~/log/execute/flight/run_manifest.json. That manifest
is the portable first-proof run record: command, Python/platform context, active
app, timing, artifact references, and validation status.
That supports an Ease of adoption score of 4.0 / 5: the public demo
works, the first routes are explicit, PyCharm is optional, installer tests are
opt-in, and the source-checkout proof now has local, fresh external macOS,
repeatable Flight cluster doctor, AI Lightning, Hugging Face, bare-metal
cluster, and VM-based cluster validation.
It is not scored higher yet because Azure, AWS, and GCP deployment validation
remains open.
After day 1: cluster proof
Cluster validation is a second milestone, not part of the newcomer proof.
Use it only after the local flight_project path has passed once.
For the repeatable two-node check, go to Cluster. That page owns:
LAN discovery and SSH prerequisites
shared cluster-share setup and sentinel checks
the Flight cluster doctor command
source-checkout and package-mode cluster validation commands
What to ignore on day 1
Skip these until the local flight_project proof works once:
cluster and SSH setup
published-package mode
notebook-first route
private or optional app repositories
IDE convenience flows
full installer test suites unless you explicitly want validation instead of the fastest first proof
The five words you need
PROJECT: where you choose the app you want to run.
ORCHESTRATE: where you install and execute it.
PIPELINE: where you inspect, generate, or replay run steps.
ANALYSIS: where you look at the result.
Worker: the isolated runtime that actually executes the app.
Common newcomer traps
Mixing package mode and source mode without being explicit: pick one first, then switch deliberately.
Trying cluster mode before a local run succeeds: local success gives you a clean baseline for later SSH debugging.
Expecting private or optional apps to appear automatically: public built-in apps live under
src/agilab/apps/builtin; extra apps usually requireAPPS_REPOSITORY/AGILAB_APPS_REPOSITORY.Running ``uvx agilab`` from the source tree: from a repository checkout, use the source commands documented in Quick-Start so you do not accidentally run the published wheel.
Assuming PyCharm is required: PyCharm mirrors are useful for debugging, but the supported first proof is independent of PyCharm.
Where to go next
Quick-Start for the exact first-proof commands.
Newcomer First-Failure Recovery if the local
flight_projectproof fails.AGILAB Demo if you want the public hosted web UI instead of a local install.
Demos if you want the public demo chooser.
agi-core Demo only if you intentionally choose the
agi-corenotebook path.Distributed Workers only after the local proof works and you are ready for SSH or multi-node execution.