Follow the evidence. Find your way in.

Welcome to CueCrux: what is this place?

A short orientation to the control plane for agent fleets

CueCrux gives agent fleets signed receipts, passport identity and enforceable policy; this section is your map to it.

How do I get my first receipt?

Install the daemon, connect an agent, watch the record begin

Install the free daemon, connect an agent over MCP, and the first state mutation mints the first signed CROWN receipt.

How do I read the record my agents create?

The console surfaces, the receipt trail, and what the chain proves

Learn what the console surfaces show and what a receipt trail proves: who acted, in what order, under which grants.

What should I explore next?

Where to go after your first receipts

A map of the catalogue and product pages: receipts, passports, RCX, governance, fleet, pricing and WikiCrux.

What is CueCrux and what does it do?

The governance control plane for AI agent fleets

CueCrux is the governance control plane for AI agent fleets: every action gets a signed receipt, every cost attributes, and revocation completes with proof.

How does CueCrux actually work?

One local binary; memory, retrieval and receipts on an append-only spine

One local binary runs memory, retrieval and receipts on an append-only spine; agents connect over MCP and every action lands on the record.

Why should I trust CueCrux?

You should not have to: every claim it makes checks out offline

Trust is not requested: receipts are signed, hash-chained and verifiable offline, and tamper rejection is a release-blocking CI gate.

What is a CROWN receipt?

A signed, hash-chained record of one action, verifiable offline

A CROWN receipt is a signed, hash-chained record of one state mutation, resolvable to who acted and verifiable offline.

Is CueCrux free or does it cost money?

The local daemon is free forever; paid tiers govern fleets

The local daemon is free forever; Pro and Governance add hosted sync and fleet controls, and metered credits mint receipts of their own.

How is CueCrux different from ChatGPT?

ChatGPT is an assistant; CueCrux governs the agents you run

ChatGPT is an assistant; CueCrux governs the agents you run on any model, with memory that benchmarks show holds up at scale.

Who is CueCrux designed for?

Teams that run agents and need a verifiable record of what those agents did

CueCrux serves platform teams running agent fleets, engineers debugging them, compliance teams answering for them, and solo developers on the free daemon.

What does CueCrux search?

Your own store, retrieved under hard token budgets

Retrieval runs over the store your agents build: local lexical and graph lanes, bring-your-own dense, hosted richer lanes, all under hard token budgets.

How do I get started with CueCrux?

Install one binary; no account, no GPU, no cloud

Install one free binary, open the local console, connect an agent over MCP, and verify your first receipt; no account, no GPU, no cloud.

Is my data safe and private with CueCrux?

Local-first by default; nothing leaves your machine unless you choose

The default deployment is local-only: your store stays on your disk, writes are passport-bound, and erasure is a governed operation with a dry-run.

What is RCX and what are its postures?

Per-request policy: a scoped capability token or a signed refusal

RCX evaluates every request and answers with a scoped capability token or a signed refusal; rollout moves through off, observe and enforce.

Can CueCrux hallucinate?

The record is replayed, not remembered

Answers about your fleet are replayed from signed, hash-chained records, not generated; fabrication breaks the chain and verification catches it.

Does tool-mediated memory actually hold up?

ScoreCrux is the open benchmark; the numbers are published

ScoreCrux, the open benchmark, shows stuffed-context recall collapsing at scale while tool-mediated memory holds 80-100% at a fraction of the tokens.

Can I share proof of what my agents did?

Export the trail; anyone can verify it offline, no account needed

Export a context and anyone can verify the receipt chain offline, without an account; tampering is rejected and the rejection path is CI-gated.

How does CueCrux help with the EU AI Act?

The compliance machinery is the same machinery everything runs on

Risk classes, human gates, hash-chained logging and provable erasure are runtime controls that double as EU AI Act evidence.

What happens when I revoke access?

A signed pull you can watch complete, agent by agent

Revocation is a signed pull with per-agent acknowledgements, ending in an attestation when zero holders remain.

Ask Surface

Questions in, cited answers out, receipts behind everything

The question-and-answer surface where every answer opens its sources and every retrieval ran under a budget.

Proof Surface

The CROWN receipt chain, rendered and verifiable offline

Where receipt chains become inspectable: signatures, hash links, refusals, and offline verification.

Watch Surface

Calm monitoring of fleet activity and evidence drift

Low-noise monitoring: what changed in your fleet's trails and your evidence base, and whether it matters.

Community Threads

Corrections and source suggestions, passport-bound and receipted

The maintenance queue where the community improves the knowledge base, with reputation earned through verified receipts.

Signals Feed

The mechanical stream of everything your fleet did

A chronological feed of receipts, refusals, gate decisions, and revocation acks, filterable and machine-consumable.

Canvas White-Label

Your brand on the surfaces, CueCrux machinery underneath

Run the receipted surfaces under your own brand and domain; branding never touches the hash chain.

Reasoning Desktop

The Crux Daemon on your machine: one binary, no account

The local-first CueCrux experience: a free CPU-only daemon, an operator console, and receipts that verify offline.

Domain Directory

A browsable, honest map of what the knowledge base covers

The catalogue of knowledge domains, with coverage shown as it is rather than as marketing would prefer.

Start here: the CueCrux doc directory

Guide

A short map of the guides, FAQs, and legal pages in this catalogue, and which one to read next.

Getting started with the Crux Daemon

Guide

One free binary, running locally in about 60 seconds. No account, CPU-only, and every state change signed from the first minute.

Rollout postures: off, observe, enforce

Guide

RCX policy rolls out in three postures. Start by watching what would be refused, then turn enforcement on with evidence in hand.

Credits, metering, and what is free

Guide

The local daemon is free forever. Paid tiers and metered credits exist for hosted heavy lifts, and every credit spend mints a signed receipt.

Operating a fleet: passports, grants, and forgetting

Guide

Identity, permission, revocation, and erasure for agent fleets: the four controls an operator actually uses, all of them receipted.

Press kit

Guide

Boilerplate, naming, and logo rules for writing about CueCrux.

Sharing proof: receipts that verify anywhere

Guide

Hand an auditor, customer, or colleague an export they can verify offline. No CueCrux servers, no trust in our UI required.

Troubleshooting the daemon

Guide

What budget trims, refusal receipts, verification failures, and slow revocations mean, and the next step for each.

Where CueCrux is heading

Guide

What ships today, what lands with the launch gate, and the one principle that does not change.

Why trust a receipt

Guide

What makes a CROWN receipt evidence rather than a log line: signatures, hash-chaining, offline verification, and identity that resolves.

What is CueCrux and what problem does it solve?

General FAQ · Platform basics

CueCrux is the governance control plane for AI agent fleets: cost attribution, receipts you can replay, and revocation that provably propagates.

How does CueCrux produce a verifiable record of what an agent fleet did?

General FAQ · Receipts and proof

Every state mutation gets a signed CROWN receipt on an append-only hash chain. The trail verifies offline, so it is evidence, not telemetry.

How is CueCrux different from agent observability dashboards?

General FAQ · Platform basics

Observability shows you what a logger asserts happened. CueCrux gives you signed, chained, offline-verifiable evidence, plus controls that act on it.

What is a CROWN receipt?

General FAQ · Receipts and proof

The signed, hash-chained record the daemon mints for every state mutation: who acted, what changed, when, and under which grants.

How does retrieval work inside the daemon?

General FAQ · The daemon

Local BM25-class lexical search plus graph fusion on CPU, dense retrieval via your own embedder, and richer hosted lanes when you want them.

What is WikiCrux?

General FAQ · Platform basics

The receipted knowledge hub, live at wiki.cuecrux.com. Cited answers open their sources, and access is governed by RCX like everything else.

Can a receipt be tampered with or forged?

General FAQ · Receipts and proof

Tampering breaks the hash chain visibly, forgery needs a passport signing key, and the tamper-rejection flow is a release-blocking CI gate.

Can I replay what my agents did?

General FAQ · Receipts and proof

Yes. Receipts are the debugging surface: replay the signed trail instead of reconstructing behaviour from logs.

Does tool-mediated memory actually beat stuffing the context window?

General FAQ · Evidence

On the open ScoreCrux benchmark: stuffed-context recall collapses at 2M-token scale while tool-mediated memory holds 80-100%.

What is a token budget and why is it mandatory?

General FAQ · The daemon

Every retrieval call carries a hard token_budget and the daemon trims to fit. It is the guard against context blowouts, and it is not optional.

Do I need an account to use CueCrux?

General FAQ · Getting started

No. The local daemon is free, needs no account, and never phones home for permission. Accounts exist only for the hosted tiers.

How does CueCrux handle GDPR and the right to be forgotten?

General FAQ · Privacy and identity

Scoped forget with a dry-run implements Article 17: preview exactly what will be erased, erase it, and keep chain integrity via hashes, not content.

How does CueCrux prevent anonymous or spoofed agent activity?

General FAQ · Privacy and identity

Every connection binds to a passport, every tool call carries it, and reputation accrues across five trust tiers from verified receipts.

How do metered credits work?

General FAQ · Pricing

Roughly £0.01 per credit, charged only for hosted heavy lifts. The local dense lane is never metered, and every spend mints a signed receipt.

Can CueCrux run a whole fleet on one shared brain?

General FAQ · Fleet and governance

Yes: shared memory on one spine, cross-session handoffs, a live work board, and governance that is identical for one agent or hundreds.

What do we hand an auditor or regulator?

General FAQ · Fleet and governance

An export of the relevant receipt trail, which they verify offline on their own machine. No access to your systems, no trust in ours.

What is the local operator console?

General FAQ · The daemon

A console at localhost:14800, shipped with the free daemon: canvas board, relation graph, ExecPlans kanban, and sessions.

How do my agents talk to the daemon?

Developer FAQ

Over MCP: 100+ tools for memory, retrieval, sessions, and plans. Every call carries a passport and a hard token_budget.

What happens when a call is not allowed?

Developer FAQ

RCX evaluates every request against short-lived capability tokens. Refusals fail closed with a signed RefusalReceipt and a reason code.

How do I verify receipts in code or CI?

Developer FAQ

corecruxctl verify-store checks a whole store; exported trails verify offline anywhere. The same flow gates our own releases.

Can I bring my own embedder for dense retrieval?

Developer FAQ

Yes: Ollama, vLLM, TEI, llama.cpp, or LiteLLM. The local dense lane fuses with lexical and graph, and is never metered.

How do I move context between sessions and machines?

Developer FAQ

Sessions, checkpoints, and cross-session handoffs cover the local case; .cruxpack export/import moves context between machines with provenance intact.

Can I drive plans, decisions, and procedures programmatically?

Developer FAQ

ExecPlans and the live work board are tool surfaces, decision records capture the why, and engrams compile procedures with provenance hashes.

Do I ever need CueCrux's cloud?

Developer FAQ

No. The daemon is local and CPU-only, and verification works offline. Hosted CoreCrux adds richer lanes and sync only if you want them.

Why does an agent fleet need a governance control plane?

Enterprise FAQ

Because agents act at machine speed and scale, and the questions that follow (what did they do, what did it cost, is that permission gone) need proof.

What does "revocation that propagates" actually mean?

Enterprise FAQ

A signed pull goes out, every holding agent acknowledges, and an attestation issues when zero holders remain. Settlement is proven, not assumed.

How do passports map to people, services, and agents?

Enterprise FAQ

One passport per actor, human or agent. Reputation accrues across five trust tiers, and any receipt resolves to who, at what tier, under which grants.

How do we reconcile immutable audit trails with data erasure?

Enterprise FAQ

Receipts commit to hashes, not content, so scoped forget erases the data while the chain keeps its integrity. Erasure itself is receipted.

How does CueCrux map to the EU AI Act?

Enterprise FAQ

Risk classes on work items, human gates that cannot auto-approve past a timeout, and hash-chained logging: engineering controls, not a legal opinion.

What does the Governance tier add over Pro?

Enterprise FAQ

Governance (£49/user·mo) adds fleet-wide attribution, policy gates, and revocation that propagates with proof. Pro (£19) covers sync and multi-device passports.

What market gap does CueCrux fill?

Investor FAQ

Enterprises are deploying agent fleets faster than they can govern them. The layer that proves, attributes, and revokes agent behaviour is missing.

How is CueCrux different from other companies in the agent space?

Investor FAQ

Evidence rather than dashboards: signed, offline-verifiable receipts as the substrate, with the compliance machinery identical to the operating machinery.

How does CueCrux monetise?

Investor FAQ

Free local daemon as the wedge; Pro at £19 and Governance at £49 per user per month; metered credits (~£0.01/Cr) on hosted heavy lifts, every spend receipted.

What makes CueCrux defensible?

Investor FAQ

Accumulated verifiable history: hash-chained receipts and passport reputation gain value in place and cannot be exported to a competitor as evidence.

How large is the opportunity?

Investor FAQ

Every organisation running agents at scale needs attribution and governance; regulated sectors need it first, on the EU AI Act's statutory timetable.

What stops a large vendor from copying CueCrux?

Investor FAQ

Bolt-on logging is not evidence. Matching CueCrux means rebuilding storage, identity, and policy around signed receipts, against conflicting incentives.

Accessibility statement

Legal & policy

Our accessibility commitments for cuecrux.com, this explore catalogue, and the local operator console, and how to report a problem.

Privacy notice

Legal & policy

How CueCrux handles personal data: local-first by design, minimal collection for hosted services, Paddle for payments, and GDPR rights including receipted erasure.

Refund policy

Legal & policy

How refunds and cancellations work for CueCrux subscriptions and credit top-ups purchased through Paddle, our merchant of record.

Terms of use

Legal & policy

The terms governing cuecrux.com, the hosted services, and the Crux Daemon software: licences, acceptable use, credits, disclaimers, and liability.

Can a fleet of agents share one memory?

Many agents, one spine: shared memory with governance attached

Many agents share one receipted memory spine: attributable writes, budgeted retrieval, handoffs, and a work board humans can see.

Can CueCrux tell me what my agents cost?

Attribution from fleet to agent to verb to receipt, nothing estimated

Every unit of spend resolves from fleet to agent to verb to receipt, with nothing estimated and every credit spend minting its own receipt.

What does the Governance tier include?

Fleet-wide attribution, policy gates, and revocation with proof

Governance, at £49 per user per month, adds fleet-wide cost attribution, policy gates, and revocation that propagates with proof.

How does CueCrux plug into my agents and tools?

MCP-native: 100+ tools any compliant agent can call

The Crux Daemon speaks the Model Context Protocol and exposes more than 100 tools for memory, retrieval, work tracking and governance, so existing agents integrate without a custom SDK.

Can I export receipts and memory?

Context export, .cruxpack portability, and attestation for outputs

Context exports verify offline and reject tampering, .cruxpack moves whole stores, and C2PA attestation is available behind a default-off flag.

Can work continue across sessions?

Sessions, checkpoints, handoffs, and procedures that carry their provenance

Sessions, checkpoints, handoffs, ExecPlans and engrams carry work between sessions with the receipts attached.

What are passports and trust tiers?

Identity that earns reputation: bind, carry, earn, answer

Every connection binds to a passport, every call carries it, and verified receipts accrue as reputation across five trust tiers.

How does CueCrux handle my private data?

Local-first by design: memory and retrieval run on your machine

The Crux Daemon stores and searches agent memory on your own hardware. Nothing leaves the machine unless you explicitly opt into hosted lanes, and every state change is signed and attributable.

What happens when stored knowledge changes?

Append-only history: corrections add, they never silently overwrite

Every state change in CueCrux is a new, signed entry on an append-only spine. You can always see what was recorded, when, by whom, and what superseded it.

Does context survive across sessions?

Sessions, checkpoints and handoffs instead of one endless chat

Work in CueCrux is organised into sessions with checkpoints and cross-session handoffs, so an agent can stop, resume, or pass work to another agent without losing the trail.

How do teams and their agents share memory?

Many agents, one shared spine, every write attributed

CueCrux gives a fleet of agents one shared memory spine: common facts, work boards and handoffs, with every write bound to a passport so nothing is anonymous.

Can I share evidence with people outside my organisation?

Export the trail; anyone can verify it offline

CueCrux evidence travels. Export a context pack, hand it over, and the recipient verifies the signatures and hash chain offline, with no account and no connection to your systems.

How do permissions work in CueCrux?

Per-request capability tokens, not standing role checkboxes

RCX replaces standing roles with short-lived capability tokens scoped to tools, tenant and tier, issued per request and verifiable offline. Refusals fail closed with a signed receipt.

How does organisation-level governance work?

Fleet-wide attribution, policy gates and revocation with proof

The Governance tier gives an organisation one view over its whole fleet: every action and every unit of spend attributed through fleet, agent, verb and receipt, policy gates on risky work, and revocation that provably propagates.

What does the audit trail look like?

Receipts are the log: signed, hash-chained, tamper-evident

CueCrux does not keep a separate audit log. The receipt chain is the log: every state mutation is signed with Ed25519 and chained on a BLAKE3 append-only spine, and tampering is detectable offline.

Where does my data live?

On your machine by default; hosted lanes are strictly opt-in

The strongest residency guarantee is not a region picker but a binary on your own hardware. The Crux Daemon keeps memory, receipts and retrieval local unless you explicitly opt into hosted services.

Healthcare Compliance

Agent fleets in HIPAA-regulated work, with proof

Run AI agents on healthcare compliance work where every touch on patient-adjacent data leaves a signed, replayable record.

Financial Regulation

Auditable agent fleets for Basel, MiFID II, and Dodd-Frank work

Deploy agents on regulatory research and reporting where every figure traces back through a signed receipt chain.

Is CueCrux suitable for legal teams?

Confidential by architecture, evidential by default

Legal teams adopting AI agents get a memory and governance layer that keeps privileged material on their own machines and produces a signed, replayable record of what the agents actually did.

Academic Research

Literature reviews with citations that resolve

Systematic review where the search strategy is receipted and every citation opens its source.

Due Diligence

Evidence packs your counterparty can verify offline

Agents assemble transaction diligence where every finding is receipted and the pack verifies without trusting you.

What does CueCrux give security teams?

Fail-closed policy, tamper-evident logs, revocation with proof

Security teams get controls built for agent fleets: per-request capability tokens that fail closed, an append-only evidence chain that exposes tampering, and revocation that propagates with per-agent acknowledgements.

How does CueCrux map to the EU AI Act?

The compliance machinery is the machinery everything runs on

CueCrux does not bolt compliance onto the side. Risk classes, human gates, hash-chained logging and provable deletion are the same mechanisms every agent action already uses, mapped to the obligations the EU AI Act describes.

Regulatory Filing

Submissions where every claim resolves to a signed source

Agent-assembled filings with receipted evidence and a human gate before anything reaches the regulator.

Evidence Review

Litigation support with a cryptographic chain of custody

Systematic review where custody is the hash chain and privileged material never leaves your machines.

Compliance Audit

Give the auditor a trail to replay, not a deck to believe

Gap analysis where findings are receipted, remediation is tracked, and the audit itself is auditable.

Policy Comparison

Cross-jurisdiction analysis with every difference cited

Agents sweep multiple jurisdictions and every cell in the comparison matrix resolves to its source provision.

Claim Verification

Check assertions against sources, then check the checker

Third-party claims evaluated against receipted evidence, with a verification trail that itself verifies offline.

Risk Assessment

Risk classes that gate what agents may do next

Assessment wired into execution: risk classes on plans and work items decide what needs a human gate.

Board Reporting

Fleet reports built from receipts, not from asking around

What the agents did, what it cost, what was refused, and what humans approved, derived from the signed spine.

Incident Investigation

Replay the signed trail instead of interviewing the logs

When an agent does something wrong, the receipt chain shows exactly what, in order, tamper-evident.

Market Research

Research agents that show their working

Competitive intelligence with provenance on every insight and token budgets on every sweep.

Patent Landscape

Prior art sweeps where the methodology is the receipt trail

Landscape and novelty analysis with a signed record of what was searched, when, and in which corpus.

Vendor Assessment

Third-party risk, assessed with evidence you can keep

Structured vendor evaluation with receipted findings, and a hard question for AI vendors: can they show a verifiable record of what their agents do?

Fraud Detection

Anomaly analysis over tamper-evident trails

Investigation findings that stand up, because the record they rest on rejects tampering by construction.

Product Launch

Ship AI features with the compliance evidence already minted

Launch readiness where risk classification, human gates, and hash-chained logging run from day one of the build.

Contract Negotiation

Benchmarking positions with evidence, not folklore

What is market gets asserted constantly and evidenced rarely; receipted benchmarks change the conversation.

ESG Reporting

Sustainability metrics with a signed trail to the source

CSRD, TCFD, and ISSB disclosures assembled by agents, with every number traceable and the assembly tamper-evident.

IP Law

Prior art and IP research with a receipted search trail

Agent-run patent and trademark research where the methodology itself is signed, timestamped, and replayable.

Employment Law

High-risk AI territory under the EU AI Act, handled as engineering

Agents in hiring and HR workflows run under risk classes, human gates, and signed refusals.

Data Privacy (GDPR)

GDPR obligations as controls the daemon actually enforces

Provable processing, scoped erasure with dry-run, and no anonymous writes, enforced by the same machinery everything runs on.

Contract Analysis

Clause extraction you can trace back to the page

Agent-run contract review where every extracted clause and risk flag carries a receipt to its source passage.

Tax Legislation

Tax research trails built to survive a revenue authority

Agent-assisted tax and transfer pricing research where positions are date-stamped against the guidance in force at the time.

Environmental Policy

ESG and disclosure work with receipts against greenwashing claims

Agents assembling CSRD, TCFD, and ISSB disclosures where every number carries a signed trail to its source.

Pharmaceutical R&D

Regulatory-grade audit trails for agents in drug development

Agents in submission prep and literature surveillance under records that are attributable, contemporaneous, and tamper-evident.

Insurance Regulation

Model governance for agent fleets under Solvency II scrutiny

Insurers run agents on claims and actuarial work with the attribution and replay that model governance expects.

Anti-Money Laundering

Screening and monitoring agents whose decisions replay

AML work where every alert disposition, screening decision, and refusal is signed and attributable.

Cybersecurity

Agent fleets in the SOC, and revocation that actually propagates

Security teams run triage and enrichment agents with signed trails, scoped credentials, and provable revocation.

Banking Compliance

PSD2-era banking work with per-request, revocable authority

Banks apply open-banking consent discipline to their own agents: scoped tokens, signed refusals, provable revocation.

Corporate Governance

Board-grade oversight of what the AI fleet actually does

Directors get attribution, gates, and decision records instead of assurances about the organisation's AI usage.

AI Regulation

The EU AI Act as engineering controls, not a legal opinion

CueCrux implements the Act's oversight, logging, and transparency demands with the same machinery everything runs on.

Antitrust & Competition

Deal-room research agents with defensible trails

Merger and conduct work where agent-gathered evidence carries provenance and stays on your infrastructure.

Supply Chain Compliance

CSDDD-grade due diligence with evidence, not attestations

Agents scale supplier due diligence across thousands of entities while every finding stays traceable.