Atlas v0.1.0 — first tagged checkpoint (reuse contract + Phase 0 prod…

… hardening + Phase 1 _user overlay auto-launch)

Phase B complete: 21 row files + 32 candidate one-pagers + integratio…

…n matrix

Phase A complete: deps standardization + manifest-driven diagram gene…

…rator

modularizes per-service configuration into services/<name>/{service.y…

…ml,compose.yml}

- replaces 1,425-line monolithic docker-compose.yml with a 52-line include:-shell that merges 19 per-service fragments
- adds services/<name>/service.yml manifests (JSON-schema-validated) with declarative source variants and 4 Python hooks for the complex cases (comfyui, weaviate, openclaw, cloud-providers)
- builds bootstrapper/services/{manifests,manifest_validator,env_assembler,hooks}.py + bootstrapper/schemas/, bootstrapper/tools/validate_fragments.py, bootstrapper/tests/ (110 tests), .github/workflows/services-lint.yml
- captures byte-equivalence baseline at bootstrapper/tests/fixtures/rendered_config_baseline.yml; full stack renders byte-identical to monolithic
- marks bootstrapper/service-configs.yml DEPRECATED but operational; bootstrapper Python refactor to read manifests is the deferred follow-up
- documents in docs/CONTRIBUTING-services.md and docs/ROADMAP.md (modularization moved to Completed)

lands LiteLLM gateway, DB-driven model catalog, and Textual wizard re…

…work; audit cleanup; env-file resync

- LiteLLM becomes the mandatory always-on LLM gateway; cloud providers
  (OpenAI / Anthropic / OpenRouter) flow through it as independent toggles
  alongside the Ollama upstream.
- public.llms is the single source of truth for the model_list. The
  bootstrapper writes a stub config; the new litellm-init container
  renders the real config.yaml from SELECT provider, name FROM
  public.llms WHERE active = true.
- New llm-catalog-init container UPSERTs bootstrapper/utils/llm_catalog.py
  into public.llms and applies wizard / .env-driven model selections on
  every docker compose up. Schema migration 05a switches the llms
  uniqueness from UNIQUE(name) to UNIQUE(provider, name).
- Wizard gains live model fetches (OpenAI/Anthropic/OpenRouter /v1/models,
  Ollama /api/tags + ollama.com/library scrape), per-provider secret +
  multiselect step pairs, unified Ollama picker with [pulled] / [library]
  badges, Cloud APIs overview block, and a session-log tee.
- Carries the canonical service key through PromptStep so the wizard's
  command preview emits the right --<key>-source flags (previously the
  preview produced invalid --llm-engine-source / --document-processor-source).
- enforce_runtime_invariants() now returns a bool; start.py halts before
  validation when the repair-pass .env write fails (no longer silently
  recoverable).
- _render_step switches to dataclasses.replace; wizard total step count
  recomputed after dynamic splice; InfoPanel.update_state propagates to
  the visible CloudApisRow; three-state cloud footer counts (ready /
  missing key / off).
- CloudProvider gains explicit key and user_models_var fields; all
  name.lower() round-tripping removed; PROVIDERS_CLOUD in
  llm-catalog-init/sync-catalog.py derived at runtime from cloud_providers.py.
- Drops the in-SQL Ollama seed (08-seed-data.sql) — llm-catalog-init is
  now the only source of truth.
- Doc sync: provider-qualified SQL examples, Python 3.10+ prerequisite,
  Ollama pre-check behaviour, OLLAMA_CUSTOM_MODELS host-side caveat,
  LITELLM_DEFAULT_MODEL fallback shape, cloud-tile to cloud-provider
  wording, removal of obsolete _live_ollama_options references.
- .env.example rebuilt as the complete inventory (258 vars, every
  auto-managed value annotated # AUTO-MANAGED and placed in its
  parent section — no more bottom-lumping on first launch).
- Other polish: .gitignore narrowed under volumes/litellm/, ollama-pull
  :ro, number-step hint when default is None, check-kong-routes symlink
  cleanup, splash-row Space-toggle guard, llm-catalog-init color in palette.
- Kong volumes/api/kong-dynamic.yml removed from git (now a generated
  artifact). Old ensure-litellm-db.sh replaced by litellm-init/scripts/init.py.