Quick-Start
If you are new to AGILab, do one thing first: run the built-in
flight_project locally from the web UI.
That first proof is:
source checkout
local install
web UI
PROJECT-> chooseflight_projectORCHESTRATE->INSTALLthenEXECUTEANALYSIS-> open the default view
If that works once, then branch into notebooks, cluster mode, or package mode. If it fails, use Newcomer First-Failure Recovery.
For architectural context, see AGILab Architecture and AGI Core Architecture.
Prerequisites
Python 3.11+ with uv installed (
curl -LsSf https://astral.sh/uv/install.sh | sh).macOS or Linux shell (use WSL2 on Windows until native support lands).
If you plan to explore remote workers later, keep SSH access for that later step; it is not needed for the first proof path.
Recommended first proof path
Use this path exactly once before trying anything broader:
Clone the repository and install the built-in apps:
git clone https://github.com/ThalesGroup/agilab.git cd agilab ./install.sh --install-apps --test-apps
Launch the web interface:
uv --preview-features extra-build-dependencies run streamlit run src/agilab/About_agilab.py
Keep the first run local and use the built-in flight demo
In the UI, use:
PROJECT-> selectsrc/agilab/apps/builtin/flight_projectORCHESTRATE-> clickINSTALL, thenEXECUTEANALYSIS-> open the default built-in view
Check the first proof outcome
You are past the newcomer hurdle when both are true:
fresh output exists under
~/log/execute/flight/you can open the default
ANALYSISview forflight_project
Only after that, branch into alternative paths
Do not switch to packaged install, notebook-first, or cluster setup before this local proof works once from end to end.
If the first proof fails
Do not broaden the problem immediately. Stay on the built-in local path and run the explicit newcomer proof first:
uv --preview-features extra-build-dependencies run python tools/newcomer_first_proof.py
If you also want the built-in app installer and seeded helper scripts checked in the same run:
uv --preview-features extra-build-dependencies run python tools/newcomer_first_proof.py --with-install
Then use Newcomer First-Failure Recovery for the five common first-run failures:
missing
uvinstaller failure
built-in app path not found
About / ORCHESTRATE startup failure
no fresh output under
~/log/execute/flight/
If you want the current public support picture before branching into other routes, use Compatibility Matrix. It makes the current validated slices explicit and separates them from routes that are documented but not the recommended newcomer proof.
Alternative install routes
Use these only if you already know why you want them.
Published package route (fastest install, less representative of the full product path):
mkdir ~/agi-workspace && cd ~/agi-workspace
uv venv
source .venv/bin/activate
uv pip install agilab
uv run agilab
Notebook-first route (best if you intentionally want agi-core before the UI):
git clone https://github.com/ThalesGroup/agilab.git
cd agilab
./install.sh --install-apps --test-apps
uv run --with jupyterlab jupyter lab examples/notebook_quickstart/agi_core_first_run.ipynb
Private apps or framework contributor setup
Only do this after the public built-in proof path is working.
Before working on private apps that depend on the public AGILab framework, initialise the pinned submodule:
git submodule update --init --recursive
Run without PyCharm (CLI wrappers)
If you work from a repository checkout and do not use PyCharm, run the
pre-generated wrappers under tools/run_configs directly from a shell.
They mirror the bundled run configurations for built-in/public apps.
List available wrappers:
find tools/run_configs -type f -name "*.sh" | sort
Examples:
bash tools/run_configs/agilab/agilab-run-dev.sh
bash tools/run_configs/apps/builtin-flight-run.sh
bash tools/run_configs/apps/builtin-flight-test-worker.sh
Optional developer workflow
If you work from a repository checkout, PyCharm and Codex CLI can speed up iteration, but they are not required to install or run AGILab.
PyCharm Professional can reuse the bundled run configurations.
Codex CLI can follow the repository guidance in
AGENTS.mdand.codex/skills.A minimal shell-only workflow remains available through
tools/run_configs.
Codex workflow
From this repository, use the shared workflow helper:
./tools/codex_workflow.sh reviewbefore larger edits./tools/codex_workflow.sh exec "short change request"./tools/codex_workflow.sh apply <task-id>
The helper resolves .external/agilab/tools/codex_workflow.sh first, then
falls back to a sibling ../agilab checkout only if you explicitly use the
legacy layout.
Cluster installs
If you want to install on a cluster, the installer must have SSH key access or
credentials with permission to deploy workers. See Cluster for the full
workflow. pycharm/setup_pycharm.py mirrors web interface run configurations to
~/log/execute/<app>/AGI_*.py so that IDE and CLI stay in sync.
Note
On a virgin workspace you do not need to hand-create
~/log/execute/<app> snippets. The installer dispatcher
(src/agilab/apps/install.py) calls _seed_example_scripts before
kicking off AGI, copying each AGI_*.py helper from
src/agilab/examples/<app>/ into ~/log/execute/<app>/ so the first
install has runnable mirrors. After that initial bootstrap, the web interface
ORCHESTRATE page re-generates the snippets on demand according to the form
inputs you provide, keeping IDE and CLI flows in sync. Field defaults are
read from each app’s per-user workspace copy
~/.agilab/apps/<app>/app_settings.toml before the form renders. That
workspace file is seeded from the versioned app_settings.toml source file
(for example <project>/app_settings.toml or <project>/src/app_settings.toml)
on first use, so update the workspace copy when you need local baselines and
update the source seed only when you intend to change the shipped defaults.
Next steps
AGILab Architecture – understand how the web interface,
agi_core,agi_envandagi_clusterfit together.Project Files Structure – explore the repository layout with an annotated tree tracked in the repo and refreshed when
docs/source/directory-structure.txtis updated.About AGILab – learn how core pages and optional page bundles fit together in the UI.
Page Bundles (apps-pages) – learn how page bundles are discovered, enabled, and launched.
Support
Support: open an issue on GitHub
License
New BSD. See License File.