v0.3.2: Chat-path latency, customer-bug batch, v0.3.1 correctness repairs

Performance (Section 0):
- Parallelize two-lane retrieval via asyncio.gather
- Warm ONNX models at startup; offload inference to executor
- WebSocket ready-poll replaces hardcoded 500ms sleep

Customer-reported bugs (Sections 0a-0i):
- Universal HTTP error handling for Ollama/LM Studio tools
- Main-model persistence via JSON file
- Stale-model 404 surfaces as actionable UI text
- ChromaDB telemetry disabled, keep_alive=24h
- Sidecar mirror-chat defaults, reload warning modal
- VRAM-aware model picker, full Ollama tag in dropdown

Laptop-testing pass (Sections 0j-0p):
- Shared-DB timestamp tolerance (desktop + core ChromaDB)
- Sidecar extract_facts/tags tolerate bare JSON array
- mirror_chat migration defers when chat model empty
- Memory Panel: flatten noun_tags JSON in endpoint
- TagCascade: over-fetch + Python filter (was broken \$contains)
- memory_bank multi-key filter wrapped in \$and
- TagService wrapper dynamic sidecar lookup (was stale closure)
- store_memory_bank awaits extract_tags_async (was coroutine drop)

Memory hygiene:
- Fact dedup on store with asymmetric scope by persistence
- Embedding passed through to avoid double-embed

UI polish:
- Chat-header sidecar picker -> read-only badge
- Error toasts for chat + sidecar parity

Test debt + deprecations (Sections 1-3):
- MemoryPanelV2 + OllamaRequiredModal test rewrites
- Pydantic V1 -> V2 migration; datetime.utcnow() -> now(UTC)
- GitHub Actions Node 24 migration

546 backend tests passing.

Author: roampal-ai

.github/workflows/tests.yml

@@ -6,6 +6,14 @@ on:
   pull_request:
     branches: [master]
 
+# v0.3.2 (Section 3): GitHub Actions runners currently execute JS actions
+# under Node.js 20, which reaches end-of-life in 2026-06 and is forcibly
+# upgraded to Node 24 in mid-2026. Opting in early here avoids a silent
+# breakage window and gives us time to bump action major versions when
+# their v5 / v6 releases ship.
+env:
+  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
+
 jobs:
   backend:
     name: Backend (pytest)

.gitignore

@@ -34,10 +34,12 @@ loopsmith.log
 roampal.log
 server*.log
 
-# IDE
+# IDE / local tooling configs (not project scope)
 .vscode/
 .idea/
 *.swp
+opencode.json
+.playwright-mcp/
 *.swo
 .claude/
 

README.md

@@ -162,7 +162,7 @@ Works with any tool-calling model via Ollama or LM Studio:
 |----------|-------------|
 | [Architecture](dev/docs/architecture.md) | 5-tier memory, knowledge graphs, technical deep-dive |
 | [Benchmarks](dev/docs/releases/v0.3.1/RELEASE_NOTES.md#benchmark-evidence) | LoCoMo evaluation, TagCascade results |
-| [Release Notes](dev/docs/releases/v0.3.1/RELEASE_NOTES.md) | Latest: TagCascade Retrieval, Sidecar LLM, ONNX CE, Two-Lane Injection |
+| [Release Notes](dev/docs/releases/v0.3.2/RELEASE_NOTES.md) | Latest (v0.3.2): chat-path latency fixes, fact dedup, universal provider error handling, shared-DB timestamp tolerance, TagCascade filter correctness |
 
 ---