docs: update README, daemon API, and figures

Author: ptlnextdoor

README.md

@@ -67,7 +67,7 @@ cp .env.example .env
 Then edit `.env`:
 
 ```ini
-# Option A — local Ollama (recommended, zero network, zero rate limits)
+# Option A — local Ollama (recommended, local-only embedding path)
 VEF_EMBEDDING_PROVIDER=ollama
 VEF_OLLAMA_EMBED_MODEL=nomic-embed-text
 

docs/daemon-api.md

@@ -196,7 +196,7 @@ Non-2xx status codes:
 | 422 | Pydantic validation error (body shape) |
 | 500 | Embedder error, ChromaDB error, generic unhandled exception |
 
-## Rate limits
+## Throughput and guardrails
 
 None. The daemon is local; there is no per-client throttling. Internally, ingestion is bounded by `VEF_CONCURRENCY` (default 10) and CPU/RAM guards.
 

docs/figures/fig5_indexing_growth.png

@@ -39,7 +39,7 @@ Same root cause as above — mismatched embedding dimensions from provider switc
 Raycast aborts `/search` after 5 s by default. The daemon shouldn't take that long; if it does:
 
 1. Check for concurrent ingest backpressure: `recall status` + `GET /progress`.
-2. Tail the log for slow embedder requests (Gemini rate limits, Ollama CPU).
+2. Tail the log for slow embedder requests (network/API latency or local Ollama CPU contention).
 3. Raise the client timeout: edit `raycast/src/lib/runner.ts` → `abortAfter(5000)`.
 
 If you're on Gemini's free tier, switching to local Ollama (`VEF_EMBEDDING_PROVIDER=ollama`) removes the network round-trip entirely.