Arcana. Memory your agents keep.
Arcana is a persistent brain for your AI agents — an entity graph, a timeline, a fact store, and semantic search, reachable from any MCP-aware tool. The difference between an agent that knows your context and one that starts every conversation from zero.
Your agents forget. Arcana is how they stop.
Every session with an AI agent starts the same way: re-explaining who you are, what you're building, and what was decided last week. The model layer is brilliant; the memory layer is missing. Arcana supplies it — one brain that every tool you use reads from and writes to.
It speaks the Model Context Protocol natively, so Claude Code, Claude Desktop, OpenClaw, and the KyberAgent runtime all plug into the same memory. What one agent learns, the next one already knows.
Four stores. One substrate.
A brain is not a vector database with a logo. Arcana keeps an entity graph, a timeline, a fact store, and embeddings in one substrate — and queries them together.
People, projects, and how they connect
Entities are first-class nodes — people, companies, projects, places — joined by typed relationships. Pin an entity and it becomes load-bearing: the brain will never prune it.
A chronological record of everything
Every event lands on a full-text-searchable timeline. Ask what happened, when it happened, and what it touched — the brain answers from the record, not from vibes.
Facts that know how sure they are
Every fact carries a source-confidence score: a user correction outranks a chat inference, which outranks an AI extraction. Agents reason over what they know versus what they suspect — and ask before assuming.
Meaning, not keywords
Every memory is embedded for meaning. Search for the onboarding docs and find them even when they're titled 'new user setup' — semantic, keyword, and metadata retrieval fused into one query.
It learns while you sleep.
A background pass runs continuously over every workspace — tagging, consolidating, linking, and pruning. You never file a memory. You wake up to a brain that organized itself.
Memory that ages like a human’s
High-relevance facts persist; low-signal trivia softens and decays. The brain stays sharp instead of silting up.
Connections you didn’t make
The sleep pass proposes relationships between entities and generates insights from patterns across memories — with provenance attached.
An identity that grows
Each cycle regenerates the agent’s living profile of you and your work — so its judgment improves, not just its recall.
It's not just search.
Traditional search finds documents with matching words. Arcana fuses semantic, keyword, and metadata retrieval across the timeline, the fact store, and the entity graph — in one query, in under 100 milliseconds.
Three retrievers, one answer
Rank fusion merges vector similarity, full-text match, and structured metadata into a single ranked result — so you find what you mean, not just what you typed.
Context injected before every turn
MCP clients can pull relevant memories into the conversation automatically. Your agent just knows — no tool calls, no prompting rituals.
Hot where it counts
Access patterns promote memories through hot, warm, and archive tiers, so the knowledge you use daily answers fastest.
Your memory is portable.
Arcana is a byte-identical cloud twin of the local KyberAgent brain. Same wire protocol, same stores, same output. Flip an agent between local and cloud without losing a thing.
The KyberAgent daemon keeps the brain local-first — private by default, fast by proximity, yours by construction.
Hosted workspaces with the identical wire protocol — reachable from every device and every MCP client, with isolated storage and its own sleep loop per workspace.
memory_backend: local | cloud — a config flip, non-destructive in both directions.
Plugs into the tools you already run
Arcana speaks MCP natively, with browser-based OAuth — sign in once, no API-key juggling. If your tool speaks MCP, it has a brain.
One endpoint, fourteen memory tools
Connect Claude Code, Claude Desktop, or any MCP client and your agent gains remember, recall, search, timeline, brain notes, and entity tools — against the same brain everywhere.
Your documents, auto-synced
OAuth connectors pull from Google Drive and Notion on a schedule — no manual uploads. Slack and GitHub are next.
Install and go
The Arcana plugin adds remember, recall, and brain-note skills to Claude Code in one step — memory as a native part of your coding sessions.
Programmatic everything
Bearer-token API access to the full brain — ingest, query, and manage workspaces from your own services.
See your knowledge as a living graph
The Arcana console renders your memories as an interactive graph — nodes, relationships, clusters of themes — so you can watch the brain think.
Free to start. No credit card.
One personal workspace. Get started.
- → 1 workspace
- → 500 memories / month
- → MCP + skills
Unlimited memories. The brain at full pace.
- → Unlimited workspaces & memories
- → All connectors
- → Hourly sleep cycles
- → Priority support
Everything in Pro, plus advanced reasoning.
- → Opus-tier reasoning
- → Custom agents
- → Dedicated Slack channel
Your agents' second brain is ready when you are
Connect a tool, save a few memories, and watch the Sleep Agent organize them. The free tier needs no card — your first agent can be remembering in minutes.