🧠
Decision Traceability
Why did I choose Clerk over building auth in-house in March? The capsule contains my full reasoning at the time + the source conversation thread.
MCP Widget · Claude Desktop / Cursor One-Click Setup
RecallRecall
Your Long-Term AI Memory Capsule
MCP One-Click Save/Retrieve: In Claude Desktop or Cursor, use ` conductor_ask mode=save_memory ` to save and ` mode=search_memory ` to retrieve. Semantic search enables finding even a six-month-old note about Clerk when querying for “user authentication方案”.
Recall is alsoConductor / NexevoBuilt-in capsule storage layer with cross-model memory injection · Available as an independent MCP widget
How It Works
From Conversation to Long-Term Memory
Not a folder, not a search box — it’s your truly usable AI memory layer.
01
AI helps you identify moments worth saving
Auto Curator scores rules (~ms per turn) after each conversation round (e.g., decide / summarize / extract keywords / generate lists / output code blocks); ambiguous cases can be re-evaluated by a lightweight LLM (model configurable via Admin Console). Strong signals automatically surface the “💡 This passage is worth saving” chip.
02
One-click encapsulation
Chat / Quorum: Hover over any message and click “Save as Memory” → opens a sheet (Haiku auto-generates the title) + automatic deduplication (if cosine similarity ≥ 0.85 within the past 7 days, prompts for merging) + tagging; Claude Desktop / Cursor / agents save memories to the same database via MCP memory_save.
03
Semantic Search + Time View
Voyage-3 1024-dimensional embedding + pgvector cosine index; the “User Authentication Solutions” query retrieves the “Clerk vs. Auth0” capsule—even with literal non-matches. The activity band’s timeline uses daily granularity + topic chips to visualize your thinking trajectory.
04
Any surface recalls history
New Chat / Agent / Quorum input field chips automatically recommend relevant capsules; Recall homepage ✨ AI recall + reverse spawn (inject into a new chat / have AI review / view full conversation); MCP memory_recall for client-side LLMs.
Capability Composition
Transform conversations into truly usable memories
Semantic search helps you find what you’re looking for · The timeline view makes it visible · Auto Curator identifies what’s worth saving · MCP enables any AI to call
Semantic Search
Voyage-3 + pgvector cosine top-K; no KEY / upstream failure triggers automatic literal fallback. Each result includes a similarity score, ensuring a consistent experience with the frontend chip / Recall homepage.
1024-dim · 87 capsules ~50ms
Timeline View (Activity Band)
Horizontal density heatmap showing capsule count per day + floating topic chip (top 5 aggregated from all tags). Click the chip to apply filtering immediately. Constellation chart becomes available once capsule count reaches ≥ 30.
Activity Density + Topic Aggregation
Auto Curator
Rule Scoring (Length / Keywords / Structure / Q&A / Politeness Penalty) — ~ms scale; strong → actively float chip; gray-area cases can be clicked 🧠 to trigger secondary judgment by the discriminator model specified in the admin console. Monitoring metrics are observable in real time.
Chinese threshold: 0.55 / 0.30
One-click MCP integration with Claude Desktop / Cursor
Three working tools: memory_save / memory_recall / quorum_review. Direct mcp-remote bridging configuration + Bearer authentication + per-user isolation. memory_recall defaults to semantic retrieval.
/docs/mcp
Typical Scenarios
What can your memory do?
Recall doesn’t think for you—it helps you remember, six months later, exactly how you thought at the time.
📚
Knowledge Accumulation
Every new concept learned from AI is distilled into a searchable capsule library. Semantic retrieval ensures you don’t waste tokens re-asking about the same topic.
🔀
Cross-project reuse
Design concept of Auth Project A, reused in Auth Project B with one click—Recall reverse-spawns and injects into a new chat.
💡
Creative Evolution
Ideas gradually take shape across multiple conversations; the Activity-band visualizes the temporal distribution of their evolution.
✏️
Study Notes
Save clear AI explanations as capsules for easy review later—no need to ask again. MCP enables Claude Desktop to access them too.
🔍
Conversation Archaeology
“What was that tool we discussed last week?” Semantic search retrieves it in a second—even if the literal text doesn’t match.
Pricing
Capacity-based subscription · Transparent, no hidden traps
Chat / Quorum / Agent are billed per token or per task; only Recall’s long-term storage is subscription-based on a monthly basis.
Free
$0
Capacity5 MB
Approximately~600 entries
Pro
$10/month
Capacity50 MB
Approximately~6,000 entries
Max
$40/month
Capacity300 MB
Approximately~36,000 entries
Single capsule limit: 1 MB (infrastructure protection). See detailed pricing for more information./pricing。
Your memory, your choice
Start free with 5 MB; semantic search + time view + MCP are all free.
Try Now