Internal management operating standard and deployment frameworks.
IMOS is the operator kit layer for institutional memory, operational discipline, attribution, and managed continuity across real work.
When the standard needs to persist, it needs a management surface. This repository exists to keep operational knowledge structured, reusable, and enforceable.
Capture the decision. Preserve the context. Keep the standard intact.
Persistent organizational cognition: decisions, verification, attribution, blockers, and institutional memory as an executable system.
Run the five-minute quickstart to start the API and verify its health response. IMOS is currently an alpha service with in-process storage; it is not yet a durable multi-tenant system.
Release status and gates: RELEASING.md.
GET /health— liveness and service identity/decisions— decision-record operations/context— structured context operations
MIT. See LICENSE.
The API persists decisions in SQLite and isolates rows by authenticated owner.
Set IMOS_DB_PATH to durable storage and IMOS_OWNER_KEY_HASHES to a JSON object
mapping owner IDs to SHA256 hashes of distinct, high-entropy operator credentials.
Send the credential in X-API-Key. Use TLS outside localhost. Health is liveness;
/ready requires configured ownership and migrated storage. Duplicate decision
IDs within an owner return 409; the server acknowledges storage only after commit.
Schema version 1 is created transactionally at startup. Unknown schema versions fail startup instead of silently rewriting data. Back up the database using SQLite backup tooling and test restoring it before upgrades. Owner credentials are not a browser identity service, and automated legacy-data migration is not implemented.
PR #16 includes useful graph and ingest exploration, but its unauthenticated path based ingestion and deletion of current license/release files prevent wholesale merging. This pilot deliberately exposes only decisions and owner-scoped summary; graph inference, arbitrary filesystem ingestion and institutional-cognition claims remain unvalidated. Tests exercise restart, duplicate handling and cross-owner denial.