Install¶
Source access
AIT is not offered as a public packaged download. These commands prepare an authorized collaborator's source checkout. They are not end-user release instructions.
AIT platform¶
From a source checkout:
Direct interception is ready immediately. AIT creates a private default
workspace, so the primary lab and Cockpit need no ait init:
For Parquet, DuckDB, plots, notebooks, and publication exports:
Initialize a named workspace only when you want engagement records, findings, or research artifacts. AIT will not silently place those artifacts in an arbitrary current directory.
AIT SDK¶
Plugin authors can install the independent ait-sdk package without installing
the platform server, database, CLI, transports, or peer tools:
The Python package is ait_sdk, requires Python 3.11 or newer, and depends only
on Pydantic 2 at runtime. It provides strict plugin and evidence contracts,
runtime-checkable plugin interfaces, compatibility gates, canonical schemas, and
bounded ait.plugin-jsonl/v1 helpers. It does not execute or sandbox plugins.
Local service boundary¶
ait server start is a local workstation service. It accepts loopback listeners
only and creates a random bearer token under the private workspace state.
The optional Compose profile is also a loopback workstation deployment; native installation is the most portable path.
Seam¶
Assay¶
Assay expects a running Seam local API when executing cases.
meshmapper¶
meshmapper reads Seam transcripts and other saved artifacts, then writes graphs and path hypotheses.
Documentation¶
For CI-style validation:
The repository workflow stages an explicit allowlist before building the public site. Publication and research material are excluded by default. Do not bypass that boundary with a manual Pages deployment.