AOI — Agent Organization Infrastructure¶
Git-native governance for coding-agent teams.
Run Codex or Claude Code normally. AOI keeps ownership, delegation, decisions, checkpoints, and verification accountable when the work becomes parallel, long-running, or evidence-sensitive.
you set the goal
|
Codex / Claude Code reasoning, tools, implementation
|
AOI client adapter + hooks lifecycle context and runtime observations
|
AOI CLI/runtime/ledger authority, ownership, evidence, recovery
|
Git + tests + build / EDA source of truth
AOI targets a recurring failure mode: delegation succeeds, several agents become busy, and nobody can later prove who owned a file, which result used the current baseline, or whether "tests passed" meant that a test actually ran.
Alpha status
AOI's lifecycle and integrity rules are tested, but AOI has not established general superiority over a strong single agent or a simpler supervisor. It is a cooperative procedural guardrail, not a security sandbox. Test it on your own workload before making it the default.
Install¶
These docs describe an unpublished ARISE Operational Alpha candidate
It is not a complete v0.5 release. Current schema-v3 hooks require reviewed local-v2 exact-wheel proof. Public schema-v1 receipts remain readable and historically upgradable but cannot newly enable current hooks; the public current-hook route is deferred. It claims neither publication nor live Codex hook trust.
For a reviewed local install, create a repository-external venv, install the
exact bundle-bound wheel with --no-index --no-deps, then run the venv's exact
aoi console launcher. codex-init needs one complete proof pair:
<venv>/bin/aoi codex-init --project-name "My Project" \
--local-artifact-bundle-file /absolute/reviewed-local-install-bundle.json \
--expected-local-artifact-bundle-sha256 '<approved-local-install-bundle-sha256>' \
--json
The local reviewed_local_install_bundle has
proof_scope=exact_local_wheel_install_only: it is neither a release nor a
promotion, and local-v2 console/bridge/legacy-launcher evidence is not v3
active-hook authority. See the full Operational Alpha quickstart.
What AOI records¶
- Authority — one Chief lease at a time, epoch-fenced, with repo-external credentials; sub-agents return bounded results and never write state.
- Ownership — explicit claims over exact file/tree locks before mutation.
- Evidence — verifications carry a category, an explicit boundary, and —
for a task to close
achieved— an assertion that the registered completion boundary was actually covered. - Routing honesty — a model-routing claim is only
verifiedwhen a hook observation matches an applied configuration binding; operator free text is recorded as a claim, never as verification. - Recovery — crash-consistent state with fail-closed recovery paths and a doctor that treats unprovable claims as errors.
Where to go next¶
- Architecture — planes, authority model, and how AOI sits beside a coding-agent client.
- Configuration — the strict
aoi.tomlschema. - Operating policy — the full governance contract.
- Resource control — governed Codex model/concurrency configuration.
- Release runbook and the helper-budget live canary.