Explore MemoryCrux
The features that turn a language model into a colleague — durable memory, sessions that carry work across runs, plans that structure it, and a console to watch it all. Not a list of tools.
Getting started
Run the daemon (free)
The Crux Daemon is source-available and runs locally — no account, no network. One command and the memory engine, MCP server, and console are live on your machine.
Connect your MCP client
Point Claude Desktop, VS Code, Cursor, or any MCP-compatible client at the built-in server. Every agent in your stack queries the same memory before it acts.
"memorycrux": {
"url": "http://localhost:14801/mcp"
}
Encode your judgment
Capture facts, decisions, and constraints — natural language in, machine-checkable boundaries out. Point the daemon at a folder with corecruxctl ingest ./docs, or let agents record what they learn as they go.
Watch it in the console
Open the console at localhost:14800 to see the fleet, the gate queue, cost, and your memory — every state mutation receipted and auditable. Upgrade to sync and govern it across machines.
http://localhost:14800
What you can do
The memory core, exposed as one MCP surface of 100+ tools. Grouped by what they do — memory leads; governance follows. The same surface on every tier; access is gated by capability tokens and passport reputation, never a crippled binary.
Memory & retrieval
A versioned fact store that remembers how it changed. Store, recall, and reconstruct what was known at any point — BM25 today, bring-your-own dense when you want it.
store_factqueryquery_factsfact_historymemory_viewFreshness & drift
Memory that knows when it is stale. Facts carry a decay horizon; the daemon flags anything past it and warns an agent before it acts on out-of-date context, then resolves contradictions into a canonical value.
memory_freshnessmemory_contradictionsmemory_consolidatememory_reverifyDecisions & constraints
Record why a decision was made, declare organisational boundaries in natural language, and check a proposed action against them before it runs. Senior judgment becomes a machine-checkable boundary.
record_decisiondeclare_constraintcheck_constraintsenrich_actionIdentity & passports
Every agent earns an Ed25519 passport with a 5-tier reputation. Sync is gated by tier, so agents must build local track record before sharing knowledge across machines.
issue_passportget_passportpassport_link_deviceresolve_principalPortability & custody
Export your whole memory as a self-certifying .cruxpack, attest outputs, forget on request (GDPR), and verify any receipt offline. Your context survives the platform disappearing.
memory_forgetoutput_attestreceipt_verifyartefact_putSessions — continuity across runs
Every agent run starts from zero unless something carries it forward. A session does: at the end of a run an agent saves a checkpoint — objective, current milestone, decisions, files touched, next action — and the next run resumes from it instead of replaying the whole conversation.
Sessions are journal-backed and survive process restarts, so continuity never depends on a chat window staying open. Archive the ones you want to keep; token usage is tracked per session.
save_sessionsession_checkpointget_sessionlist_sessionsarchive_sessionExecPlans — work that spans sessions
Longer work is structured as an ExecPlan: a milestone-by-milestone document with a progress checklist and a decision log that records why choices changed mid-flight. Agents pick up at the current milestone across sessions — the plan is the shared frame that outlives any one run.
The daemon projects plans into a live work board with dependency and supersession links, and attributes token burn back to the plan that caused it — so you can see which work cost what.
list_workcreate_workupdate_work_statecoord_announcestatus_feedThe console — watch your fleet think
The daemon ships a local operator console at localhost:14800 — no account, no cloud. It reads the daemon's real data and every state mutation carries a signed receipt. Five surfaces, one shell, three themes (glass, dark, light).
Overwatch
The needs-you gate queue, the live fleet of sessions, and a rolling activity ticker — everything waiting on a human, in one view.
Canvas
A size-adaptive tile board that composes to your viewport, and a real-edge relation graph over projects, work, gates, passports, and sessions.
Documents
Your corpus as a reader — every passage carries its source and a verifiable CROWN receipt.
Explorer
Search the corpus — local BM25 or the mediated engine — with real scores and provenance.
Meters
Cost and token burn, attributed back to the ExecPlan and the agent that spent it.
The six questions an agent should be able to answer
Identity is one answer of six. An agent arriving in a new organisation should know who it is, who it works for, what it is here to do, what it must not do, how the work is usually done, and what changed since last time. The passport anchors the first; the rest plug in alongside it.
| Layer | Question | Source |
|---|---|---|
| Identity | Who am I to this platform? | Passport |
| Operator | Who do I work for and how do they work? | OperatorProfile |
| Directive | What am I here to do, specifically? | Task context / role / resume intent |
| Rules | What must I not do? | Constraints + Shield grants |
| Playbook | How is this work typically done here? | Agent Skills + Operating Rhythms |
| Continuity | What happened since last time? | resume_session + domain changelog |
See it running
Install the Crux daemon locally for free and start building durable, inspectable memory immediately.