The containers now read the unchanged .env of the matching systemd instance
(/home/llmadmin/mindnet_dev/.env, /home/llmadmin/mindnet/.env) and use the
same Qdrant, Ollama, vault and collection prefix, so both can be compared
side by side. Only container-specific values are overridden: host service
addresses (host.docker.internal instead of 127.0.0.1) and the read-only vault.
- deploy/appenv.py derives the vault mount and prefix from that .env and
rejects application keys in the container-only file.
- Remove the separate dev Qdrant, the synthetic test vault and the import step.
- UI binds 0.0.0.0 like systemd; API stays loopback by default.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- .gitea/workflows/deploy-dev.yml: prepares the host (synthetic test vault,
dev.env from the example on first run, never overwritten), then runs
deploy/deploy.sh dev. Prod workflow stays an inactive template.
- deploy/testvault: synthetic notes from scripts/make_test_vault.py plus
synthetic edge_vocabulary.md/graph_schema.md (vault sentinel).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- preflight --live: Ollama/Qdrant reachable from the container, required
Ollama models present, 768 dimensions; clear abort instead of traceback.
Static check rejects unset or container-local service URLs.
- deploy/smoke.py: non-writing post-start check via published ports.
- deploy/deploy.sh: test (no network), build, preflight, up --wait, smoke;
prod rolls back to the last good image tag.
- Workflow templates deploy-dev.yml (push develop) and deploy-prod.yml
(push main) following the Kansho pattern; not under .gitea/workflows.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Add a Docker image and separate Compose projects for development and
production, based on the stable main behaviour without the privacy
gateway. Nothing is wired into the existing systemd deploy workflow.
- One non-root image (api | ui | preflight), base image pinned by
digest, dependencies from a hash-locked universal lock without torch
and sentence-transformers (MiniLM is not on the runtime path)
- Preflight aborts on: VECTOR_DIM != 768 or non-nomic embedding
profile, implicit or cross-env collection prefix, cloud keys without
explicit egress acknowledgement, missing/empty or writable vault mount
- Vault is a read-only NAS bind mount without host-path autocreation;
healthcheck turns unhealthy when the vault sentinel disappears
- Dev runs its own Qdrant; prod reuses the existing host instance
- API published on loopback only; read-only rootfs, cap_drop ALL,
no-new-privileges
- Dev workflow kept as an inactive template outside .gitea/workflows
- Docs: operations guide and migration plan with release gates, test
baseline classification, R2/R3 and privacy gateway requirements
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Introduce a new method for persisting rejected edges for audit purposes, enhancing traceability and validation logic. Update the decision engine to utilize a generic fallback template for improved error handling during LLM validation. Revise documentation across multiple files to reflect the new versioning, context, and features related to Phase 3 validation, including automatic mirror edges and note-scope zones. This update ensures better graph integrity and validation accuracy in the ingestion process.