- **`LLMProvider`** (ABC) — minimal `complete(messages, max_tokens, temperature) -> str` interface shared by all providers

- **`ClaudeProvider`** — Anthropic Claude via `anthropic` SDK. Default: `claude-haiku-4-5-20251001`. Separates `system` from conversation messages per Anthropic API spec.

- **`OpenAIProvider`** — OpenAI Chat Completions API and any compatible endpoint (Groq, Mistral, Together AI, vLLM, LM Studio). `json_mode=True` sets `response_format=json_object`.

- **`OllamaProvider`** — Ollama local server. Subclass of `OpenAIProvider`; default `base_url="http://localhost:11434/v1"`. No API key required.

- **`GeminiProvider`** — Google Gemini via `google-genai` SDK. Uses `response_mime_type="application/json"` for native JSON mode.

- All providers default to `temperature=0.0` for deterministic JSON output.

- **`ContextEngine`** — self-aligned ingestion pipeline that wraps `DB` with LLM-powered classification.

- **`ingest(text, hint) -> int`** — 10-step pipeline: embed → sample context → LLM classify → apply hint → store → link → episode → watch triggers → contradiction check → auto-save.

- **`ingest_batch(texts, hints) -> list[int]`** — batch ingestion helper.

- LLM JSON schema: `{entity_type, importance, confidence, ttl, namespace, episode_id, suggested_links}`

- **`_heuristic_classify(text)`** — built-in keyword-based fallback; activated when `provider=None` or LLM fails. Fully offline, zero latency.

- 5-stage JSON extraction — direct parse → strip code fences → balanced brace scan → trailing comma repair → heuristic fallback. Robust on small/local models.

- Node ID: SHA-256 of `text[:200] + timestamp + PID` mod 2^50.

- Integrates with `WatchManager`, `EpisodeManager`, `ContradictionDetector` (all optional).

- `LLMProvider`, `ClaudeProvider`, `OpenAIProvider`, `OllamaProvider`, `GeminiProvider`, `ContextEngine`

- `examples/context_engine_demo.py` — auto-detects provider from env vars (Claude → OpenAI → Gemini → Ollama → heuristic), ingests 6 records, runs semantic search and context chain.

- `feather_db.__version__` → `"0.7.0"`

- `README.md` — updated to v0.7.0 architecture; added Self-Aligned Context Engine, LLM agent connectors, MCP server, LangChain/LlamaIndex sections.