Control plane
Loading…
Governance coverage
How much of the workforce agnt8x actually governs, by recorded posture.
Containment
What can be stopped from here today.
Reliability
Agent actions and outcomes in the selected window.
Workforce
Compliance findings
What the review agents found, counted from every file they assessed. Test runs are left out. Click a broker group to see its brokers; click a broker to see their files.
Loading…
Pinned findings
Views pinned from MANAGE, Intelligence, Loop findings. Recomputed now from what each loop published.
Loading…
Model supply chain
Which providers and models the workforce depends on.
Workforce
Spend and caps
Model cost in the selected window, and agents over their monthly cap.
Workflow loops and escalations
Governed work, and how often a loop handed a decision back to a human.
Workforce
Supervision map
Which agents route work through another agent.
Key risk indicators
The vertical mark is the appetite threshold. These are agnt8x defaults until your board sets its own.
Agent register
Accountability, alignment and pre-deployment testing for every agent. Click a row to open its Passport.
SENTINEL alerts
The most recent flagged tasks, newest first.
Workforce, last 10
Not connected yet
Capabilities this view will show once their data exists. Listed so nothing here reads as "all clear" when it has simply not been measured.
Strategy
Who your competitors are, what they have launched, and how your public offer compares. Research is suggested, a person confirms what to track, and every item links to the page it came from. Scanned weekly, or now with Refresh.
Loading…
Cost of the workforce
Usage spend in the selected window by meter, and per day. Platform costs are in USD.
Loading…
Unit economics
What a unit of work costs. Shown only where the work was recorded.
Loading…
Value against cost
Monthly basis: the last 30 days of usage plus the plan fee, against value modelled from role benchmarks. Modelled value is not booked savings.
Loading…
Spend controls
Model spend caps, payment budgets, and the spend authority each agent holds.
Loading…
Agent payments
What agents paid in the selected window, from the payment records. Amounts are never added across currencies.
Loading…
Not connected yet
Finance figures this view will show once their data exists, listed so nothing here reads as complete when it has not been measured.
ISO/IEC 42001
Every agent agnt8x runs, and the organisation records behind them, tested against the standard from live platform data.
Loading…
Workforce/Overview
Workforce
Every agent your organisation has hired, what it is doing and what it is worth.
⏱️
Add card
Pilot trial
Add a card to stay on Pilot, or you'll be moved to Solo at day 30. Solo keeps your oldest agent free forever.
Pilot organisation
Your agents are onboarding. Track their progress, email them tasks, invite to meetings.
Annual Savings
—
Net $ saved per year vs equivalent human hire
Annual Value Generated
—
Labour value across your workforce
Annual Platform Cost
—
Subscription + LLM + memory
Agents Hired
0
of your tier limit
Tasks Completed
0
All time · agntx8 task ledger
Avg Alignment Score
—
SENTINEL monitors continuously
Agent Activity Feed
Every autonomous tool call by your agents — live audit trail
Loading activity...
Loading messages...
My Agent Workforce
Agent
Role / Department
Status
Trust Tier
Tasks
Alignment
Value Trajectory — Agent P&L
Updates as agents learn and improve
Agent Workflows — CONDUCTOR
🔀
No workflows yet
Hire at least two agents to build a workflow. Chain them together — handoff triggers, shared context, fallback rules.
✓ CONDUCTOR is live
Build multi-agent collaboration flows in CONDUCTOR — drag agents onto the canvas, connect them, and go live. Flows saved here sync automatically.
Agent payments
Services and panels are lit by flipping GP_FLAGS ('soon' → 'live') in the payments module as each backend piece lands. Add ?flags=live to the URL for a full preview. Live panels obey the honest-zeros rule: real data only.
💳
agnt8x approves. Your issuer pays. Every payment is evidenced.
Payment services built on the Agent Passport: for companies whose agents pay, and for payment companies whose merchants get paid. agnt8x never holds your money: it holds the identity, the authority and the evidence. EightX runs the agent payment services on its own suppliers first.
0
Payment actions (30d)
0
Active mandates
0
Pending approvals
0
KYA certificates issued
Payment services
0
LIVE
Agent payments
Know Your Agent
Identity · signed certificate
A signed certificate your agents carry to prove who they are, who owns them and what they may do.
- Company and accountable person verified
- Tools, data classes and spend limits attested
- Anyone can check it; revoked with the Passport
KYA-1KYA-2KYA-3
Fee per verification
1
LIVE · SANDBOX
Agent payments
Governed enterprise spend
Card · Mastercard, Rain next
No card exists until KYA, mandate and budget pass. Each card is sized to one payment and locked to the merchant.
- Agent asks: supplier, amount, task
- Mandate and budget checked; a person approves above the threshold
- Virtual card issued, controls read back
MastercardSingle-useCard per supplier
Per-agent fee
2
LIVE · EMAIL
Agent payments
Agent-run supplier payments
Invoice · certified workflow loop
Email an invoice to the agent. It reads it, checks it, asks the right person, then pays by card.
- Invoice read by its labels; nothing guessed
- Supplier, caps and budget checked; approver emailed
- Card for exactly that invoice; evidence kept
Email intakeMastercardXero next
Platform fee
3
LIVE · SANDBOX
Agent payments
Agents buying their own tools
Model credits · software · data
Email the agent an order. It checks the vendor and its own budget, then buys on a card that only works at software merchants.
- Order read; vendor checked against the mandate
- Per-agent budget, set on this page, checked first
- Card locked to software merchants (MCC 5734); auto-reload and x402 in pilot
AnthropicOpenAIx402
Governance fee per agent
4
PILOT
For payment companies
Payment operations agents
Merchants · ATM fleet · loop pack
Certified loops run merchant and ATM operations on your own data, with your team approving what matters.
- Onboarding checks, troubleshooting, chargeback packs
- ATM cash forecasting, fault triage, dispatch
- Settlement reconciliation and site statements
Cash forecastingDisputesReconciliation
Monthly fee per market
5
DESIGN PARTNER
For payment companies
Cross-border USDC settlement
Payouts · USDC via Circle
Settle merchants across borders in minutes. Funds stay with you or Circle; agents check, release and reconcile.
- Settlement file in, one line per merchant
- Every payee screened; batch checked against its mandate
- USDC payouts instructed and reconciled
Circle USDCSanctionsTravel Rule
Basis points on volume
6
DESIGN PARTNER
For payment companies
Agent payment acceptance and escrow
Accept · verify the agent · escrow
Your merchants accept payments from other companies’ agents safely: the agent is verified, the money held until delivery.
- Buying agent presents its KYA certificate
- agnt8x checks it live (available now)
- Escrow by you or Circle; release on delivery
KYA verifyEscrowDelivery events
Per-transaction fee
Your payment setup
Rails and accounts
Every rail plugs into the same three objects: Passport, Mandate and the governed broker. Mandate limits are written onto the rail; they are never set by hand.
Rain VisaAWAITING RAIN
CARD ISSUER · EIGHTX ACCOUNT
Scoped and single-use cards funded from our own USDC. Software access to create, cap and cancel cards requested from Rain.
Mastercard Agent PaySANDBOX
CARD NETWORK
Agent tokens enrolled against the Passport, scope taken from the mandate, revocation notices back to agnt8x.
NPP / Osko (Azupay)SIMULATOR
REAL-TIME BANK TRANSFER · AUD
Pays an approved payee's PayID or BSB and account in seconds. Final once sent, so every control is checked first. Azupay UAT keys next.
BECS direct entryABA FILE
BULK PAYMENTS · PAYROLL AND SUPPLIERS
Many payments in one ABA file for the bank. Every line checked like a single payment; the batch approved as one.
XeroLIVE
AP / AR LEDGER
Bills in for supplier payments; paid status and payment reference posted back.
USDC treasuryPLANNED
STABLECOIN · OWN FUNDS
Funds the card and pays suppliers that take USDC, to approved addresses only.
Circle USDCPLANNED
PAYOUTS · CLIENT-HELD FUNDS
agnt8x instructs USDC payouts and escrow release from the client's own Circle account. The funds never touch agnt8x.
x402 walletsPLANNED
PAY-PER-CALL
Passport-bound wallets for APIs and data that charge per request, capped by the mandate.
Client bank gatewayPER ENGAGEMENT
BANK RAIL · CLIENT LICENCE
The agent instructs the client's own gateway. The bank's licence carries the money; agnt8x carries the mandate and the evidence.
Controls and evidence
Payment Mandates
Scoped spend authority on the Passport, and the source of truth for every rail. Mirrored onto Rain card limits and Mastercard token scope.
| Agent | Per-txn cap | Daily cap | Velocity | Counterparties | Approval > | Currency | Expiry | Status | |
|---|---|---|---|---|---|---|---|---|---|
| No mandates yet. Create the first mandate to give an agent scoped payment authority. | |||||||||
Budgets
How much an agent, or the whole organisation, may spend over the last 24 hours or 30 days. Checked before every payment, with spend already held for issued cards counted. Owner or Admin.
A tag counts only the spend a loop tags with it: the Buy a tool loop tags every purchase “tools”. Setting a budget for the same scope, period and currency replaces the old one.
| Applies to | Over | Cap | Spent | Left | |
|---|---|---|---|---|---|
| No budgets. Without one, only the mandate caps apply. | |||||
Agent pays an invoice
An agent asks to pay one invoice. agnt8x checks its mandate and budgets, waits for a person where the mandate says so, reserves the spend, then pays: by a Mastercard virtual card sized to the invoice, with the mandate written into its controls, or by an NPP bank transfer to an approved payee, which settles in seconds.
Checking the Mastercard connection…
Checking the NPP connection…
The agent pays only the approved payee's bank details below. Details on the invoice are compared with them and refused if they differ.
| Requested | Agent | Supplier | Invoice | Amount | Status | Card / transfer | |
|---|---|---|---|---|---|---|---|
| No payment requests yet. | |||||||
Approved bank payees
An agent pays by bank transfer (NPP or direct entry) only to a supplier or employee a person has approved here, at these bank details. agnt8x checks the details with the rail before saving them. Owner or Admin only.
| Payee | Pays to | Checked | |
|---|---|---|---|
| No approved payees yet. | |||
Bulk payments: payroll and supplier runs (BECS direct entry)
Many payments in one ABA file for your bank. Every line is checked exactly like a single payment (mandate, budget, KYA, approved payee and its bank details); a line that fails is left out with its reason. A person who did not ask approves the batch as one, every line is checked again and reserved, and agnt8x writes the ABA file. Your bank processes it overnight.
Direct entry settings (the paying account)
| Created | Agent | Kind | Process on | Lines | Total | Status | |
|---|---|---|---|---|---|---|---|
| No bulk payments yet. | |||||||
Payment activity
Every payment request, with its mandate result, approval and settlement. Real data only.
| Time | Agent | Tool / Rail | Counterparty | Amount | Mandate check | Approval | Status |
|---|---|---|---|---|---|---|---|
| No payment actions recorded yet in this window. | |||||||
Audit pack
One evidence export per agent or loop per period: every request, mandate result, approver, card reference and settlement.
KYA certificates
Your agents' certificates, and the live check a counterparty calls before trusting one. Anyone can check a certificate; no agnt8x account is needed.
| Agent | Level | Passport | Expires | Status | |
|---|---|---|---|---|---|
| No KYA certificates yet. An independent agnt8x KYA verifier issues them. | |||||
GET https://api.eightx.app/api/v1/passport/:id/attestation
POST https://api.eightx.app/api/v1/kya/verify {"certificate": "<jwt>"}
POST https://api.eightx.app/api/v1/kya/verify {"certificate": "<jwt>"}
passport_statuscertificate_statuskya_levelownermandate_summarycertificate_expiresrevoked_at
Workflow loops
♾️
Workflow loops with identity
Every loop is bound to an Agent Passport, and that is what makes it safe to run unattended: authority is scoped (the loop can only spend what its mandate allows, with counterparties it is allowed), accountability is built in (every run writes evidence as it happens, attributable to a named worker), and control is instant (one action revokes everything the loop can touch). Create a loop two ways. Option A: send any agent your workflow document — it asks the standard intake questions (repeated task? trigger and frequency? name? fail-safes? compliance and controls? cost today?), packages every connected behaviour it read out of the document under one loop title, and registers it here for your approval. Option B: click + Teach a loop and complete the same intake yourself. The sequence is fixed: approve every step in TRAINING, test the loop in CRUCIBLE, then install it here to arm its schedule. A loop with unapproved steps cannot be tested, run or scheduled.
0
Installed loops
0
Runs (30d)
—
Success rate
—
Avg cost / run (EXC)
0
Exceptions open
Loop Registry
Every standing process, its owner Passport, its trigger, its mandate binding and its cost per outcome against your baseline.
No loops registered yet. Teach the first one: send an agent your SOP or workflow document. The agent runs the intake, PRAXIS packages every connected behaviour under one loop title, and you approve the whole package here.
Assurance & evidence
What each loop has proved, and the record of every run. A model certificate is statistical. A control certificate is exact: every test case, correct, every time.
Select a workflow loop.
Outputs
Everything your agents have assessed. Start with what needs you, and search the rest when someone asks what was said about a file.
Loading…
Needs a reviewer
Controls that could not be evaluated because their evidence was missing (input absent), or that need a person's judgement (reviewer to determine). Assign an item to yourself so others know you have it, check the file, then close it with a decision and a note. Closed items stay on the record, and each file's Outputs card shows how many of its gaps are closed and by whom.
Loading…
Proposed processes
When you send an agent a procedure (an SOP), it proposes how to run it as a loop: which steps are done by code and which need a judgement. Each loop's latest proposal is shown here, with its earlier drafts underneath. Review the proposal in TRAINING; once approved it becomes the loop you test and install.
Loading…
Document formats
What this loop can read from a document, and the patterns waiting for you to confirm. When a new layout leaves a figure blank, pick that figure below and press Look for new formats: a pattern is proposed from recent files, checked against them and against every file the loop already reads, and shown to you with the text it found. Approving it takes effect on the next run, with no release and no model involved in reading the figure.
Loop Optimiser — self-improving cost engine
The same way CRUCIBLE self-heals its test batteries, the optimiser continuously assesses every installed loop for inefficiency — starting with the model mix. Simple steps (intake, extraction, formatting, routing) run on super-cheap small models; frontier models are reserved for the steps that genuinely need reasoning. Every recommendation is verified by CRUCIBLE replay against historic runs before you can apply it: same outcome, or it is not offered.
ANALYSEper-step model + token telemetry→
PROPOSEcheaper mix via PRISM routing→
VERIFYCRUCIBLE replay · same outcome→
APPLYone click · reversible
| Loop | Current model mix | Cost / run (EXC) | Recommended mix | Projected cost / run | Saving / run | Monthly saving | Verification | |
|---|---|---|---|---|---|---|---|---|
| Recommendations appear once loops have recorded runs. Example of what you will see here: "Loop A costs 42 EXC end to end on Haiku 4.5 + Opus 4.8. Route the intake and extraction steps to a small model and keep Opus 4.8 for the reasoning step: verified same outcome at 17 EXC per run — 60% saving." | ||||||||
Run History
Every execution as a named, numbered, priceable run (agntx8_loop_runs). What today logs as individual actions becomes accountable runs.
| Run | Loop | Trigger | Started | Duration | Steps ok / total | EXC cost | Exceptions | Approvals | Status |
|---|---|---|---|---|---|---|---|---|---|
| No runs recorded in this window yet. Press Run now on a loop to create the first one. | |||||||||
Triggers
How loops fire. Conversation-initiated channels are live today; standing triggers land with the trigger service. Coming soon.
Chat / Email / VoiceLIVE
Conversation-initiated. The agent applies its learned loop when work arrives on any channel.
Meeting webhookLIVE
Fire-and-forget deliverable runs from meetings, already in production.
ScheduleCOMING SOON
Cron-style standing runs: month-end close, weekend sweeps, corridor timing.
Generic webhookCOMING SOON
Any external system fires a loop with a signed call.
Document arrivalCOMING SOON
Invoice lands in the inbox or drive; the invoice-to-pay loop starts.
Ledger eventCOMING SOON
EXC balance thresholds and budget events fire treasury loops.
x402 requestPLANNED
An inbound machine payment request triggers a governed fulfilment loop.
The Structural Spine
Learned behaviour is probabilistic; the spine is not. For payment-capable loops these steps are enforced by the platform, fail closed, regardless of model behaviour on the day.
TRIGGER→
MANDATE CHECKfails closed→
SCREENINGsanctions / AML tools→
EXECUTElearned behaviour (PRAXIS)→
EVIDENCE WRITESIGNAL / VERITAS→
OUTCOMEmetered · priced
Dark steps are structural: injected mandatorily into every payment loop scaffold. The client's bespoke workflow snaps into the EXECUTE step.
Cost per Outcome vs Your Baseline
agnt8x already meters its side of every task in EXC. Add your human and systems baseline per loop and MANAGE renders the delta CFOs ask for. Honest zeros: real data only.
—
agnt8x cost / outcome
—
Baseline cost / outcome
—
Saving per outcome
—
Annualised saving
Create Payment Mandate
Scoped spend authority on the agent's Passport. Every payment loop checks it before any card or transfer exists, and anything it cannot check is refused. Daily means the last 24 hours and monthly the last 30 days. Creating a mandate for an agent that already has one replaces it.
How the agent may pay. A rail not ticked here is refused.
Payment request
Bulk payment
Certify an agent (KYA)
You are certifying as an independent agnt8x KYA verifier. You cannot certify an agent you own, hired or are accountable for, or one in an organisation you belong to or belonged to. What you record as evidence is kept with the certificate. The accountable person appears in the certificate by role only.
Teach a Loop
Workflows are bespoke, so the product is the employee who learns yours. Your agents run this intake by default: whenever an agent receives a process document or spots a repeated task, it asks the standard questions below, then registers the loop here for your approval.
- Send the document. Open your agent's chat or email it your SOP / workflow document.
- The agent asks the standard intake questions: is this a repeated task or workflow loop? What is the trigger and frequency? What should the loop be called? What are the fail-safe items? What compliance, regulations and controls apply? What does this workflow cost you today?
- PRAXIS decomposes the answers and the document into behavioural lessons: standing rules the agent will follow. Nothing activates without employer approval in Memory (ECHO / PRAXIS).
- The loop registers here with its mandate, spine and cost-per-outcome tracking attached, and appears on the CONDUCTOR canvas for editing.
Assign loop to another agent
The loop keeps one owner (the accountable Passport); assigned agents can also run it under the same mandate and spine.
What is this document?
This project has agents attached, so this document is something an agent could learn from. Tell us what it is.
Choose the agent before picking what the document is.
Suggested: this document is titled as an SOP, so it is most likely a workflow loop.
Workflow loop intake
The agent read the document and filled in what it actually says. Anything it could not find is blank rather than guessed. Correct anything wrong, fill the gaps, and it will break the process into steps for your approval.
Reading the document\u2026
Test loop in CRUCIBLE
A loop runs unattended with real authority, so it is tested like one. Every scenario in this battery is written from this loop's own fail-safes, compliance controls and approved behaviours, not from a generic template. A failed scenario proposes a corrective behaviour back into the loop.
Loading\u2026
Schedule loop
Run this loop at a set time, or on a recurring frequency. The loop must be approved (installed) for the scheduler to fire it.
Edit workflow loop
Update the loop's details. Changes apply to the next run.
Approved steps
The executable sequence signed off in TRAINING. Editing a step means retesting the loop, so revisions go back through the agent rather than being changed here.
Loading…
CRUCIBLE
Pre-deployment testing
CRUCIBLE is the examiner in the agent lifecycle. It runs a battery of edge-case, ambiguity, adversarial-authority and graceful-failure scenarios against an agent in sandbox — every action intercepted, nothing real fired — and scores reliability before the agent is trusted to go live. A failed critical adversarial scenario caps the score.
Test batteries — reusable starting points you can open, edit, clone and run.
Loading battery catalogue…
Measured accuracy
A set of real files with the answer a human got right for each one. The agent learns
from most of them and is scored on the rest, which it never sees while learning.
Reliability asks whether an agent holds up under pressure. This asks whether it
reaches the right answer on ordinary work. They are different questions and both
appear on the Passport.
Monitoring
🛡
Ontological Alignment Monitor
Real-time governance · BCG deviation detection · audit trail
0
Active alerts (7d)
0
Tasks monitored (30d)
—
Avg alignment score
v1 deterministic scoring active · LLM-as-judge against BCG anchors lands in Phase 2 once Studio is generally available
How alignment scoring works
SENTINEL scores every completed task from 0 to 100 on whether the agent stayed within its role's declared bounds. Each agent's score below is the rolling average of its last 30 scored tasks, so it reflects sustained behaviour rather than any single task.
85-100
Healthy
Operating in-bounds. No action needed.
70-84
Acceptable, watch
Above the flag threshold, so not an alert. Worth watching if the 7-day trend is sliding. A 75 is fine; a 75 that fell from 95 is the signal.
Below 70
Flagged, investigate
The agent breached one or more bounds. Every flagged task appears in Recent alerts below.
What the score measures
Completion 30Finished the task cleanly, not an iteration-limit or fallback reply.
Iterations 20Reached the answer within its reasoning budget, leaving headroom.
Memory 15Consulted its own knowledge before answering.
Scope 20No tool errors. Stayed inside its declared capabilities.
Authority 15Escalated only when warranted and respected its decision limits.
How to read it & where to resolve
Is one low task a problem? Not on its own. The rolling average absorbs noise. A sustained downward trend, or repeated flags on the same agent, is the real signal.
Find the issue: Recent alerts below lists every flagged task with its reason. Click any agent name to open its history and the per-task breakdown showing exactly which component dropped.
Resolve it: coach the behaviour in PRAXIS (Train & Build), then re-test in CRUCIBLE before redeploying. The breakdown tells you which bound was breached, so you know what to fix.
Per-agent alignment
Loading…
Agent
Score
7d trend
Tasks (7d)
Flagged (30d)
Last alert
Recent alerts
Loading…
SIGNAL: export to your observability stack
Stream SENTINEL alignment and CRUCIBLE reliability metrics into Datadog, Grafana, Splunk, or any OpenTelemetry collector. Your existing monitoring just starts receiving agnt8x governance data, with no agnt8x dashboard required.
Scrape endpoint
https://api.eightx.app/api/v1/signal/metrics
Prometheus exposition by default. Append
?format=otlp for OpenTelemetry OTLP/JSON.Scrape keys
Collectors authenticate with a scrape key, sent as the
X-Signal-Key header. Generate one per stack and revoke anytime. The full key is shown only once at creation.Loading keys…
Example Prometheus collector config
scrape_configs:
- job_name: agnt8x-signal
scheme: https
metrics_path: /api/v1/signal/metrics
static_configs:
- targets: ['api.eightx.app']
http_headers:
X-Signal-Key:
values: ['YOUR_SCRAPE_KEY']
Send loop runs to Honeycomb
Every workflow loop run arrives in your Honeycomb as an OpenTelemetry trace: the run is the top span, with one span under it for each step and each model call. Runs are sent once a minute, each one once.
What is sent: ids, loop and step titles, agent name, statuses, ratings, counts, durations, models, tokens, costs and residency.
Never sent: document text, file names, run summaries, step inputs or outputs, error messages, or the value a check read. A step's failure reason is cut before any quoted or bracketed detail.
Honeycomb ingest key
Stored encrypted. agnt8x never shows it again, only its last four characters.
Dataset (service name)
Honeycomb files the traces under this name.
Region
The one your Honeycomb team was created in.
Export
Switching on also sends runs from the previous 24 hours.
Status
Loading…
Intelligence
The nerve centre of your self-improving workforce. Three lenses over everything your agents do: Activity is what happened, Improvement Loops is what should change, and Token Analytics is what it costs and where to save. Nothing here is applied automatically: every proposed change is yours to approve.
Loading activity…
Helix — your standing improvement engine
Helix is a meta-agent that continuously reads every action, every SENTINEL alignment score, every loop run, every token receipt and every PRAXIS lesson across your entire workforce, and files improvement proposals in real time. It closes the AI loop: sense, decide, act, gate, learn, then feed what it learned back in. You stay in control: Helix proposes, you approve, and every approved change ships through the same governed rails (Passport, CRUCIBLE test, audit trail) as any other change.
SENSORS / DATA→
POLICY LAYER→
TOOL LAYER→
QUALITY GATES→
LEARNING↶
—
Open proposals
—
Approved this month
—
Projected monthly value
—
Avg confidence
What Helix looks for
Cost cheaper model mix at equal quality
Quality low-alignment steps to re-anchor
Reliability repeated failures and retries
Coverage manual glue to codify into a loop
Compliance policy gaps and control drift
Speed slow steps to parallelise or cache
Proposal queue
Loading proposals…
Forecasts extrapolate the current daily burn. 1 EXC = $0.01.
Spend per day
What your agents cost each day in US dollars. Token spend is the model providers' cost. Memory spend is each memory write, read, log entry and day of storage at your organisation's configured memory prices.
All agents
Loading…
—
Spend this period (EXC)
—
Spend this period (USD)
—
Forecast monthly (EXC)
—
Forecast annual (EXC)
Spend by agent
Every agent's token volume, cost and monthly cap. "Cost / outcome" ties spend to completed tasks, not just calls, so a cheap agent that does nothing is not mistaken for efficiency. Agents forecast to breach their cap are flagged.
| Agent | Calls | Tokens | Cost (EXC) | Cost / outcome | Monthly cap | Forecast vs cap | Top model |
|---|---|---|---|---|---|---|---|
| Loading… | |||||||
What-if simulator
Model prices are live from PRISM. Swap a model across the workforce and see the projected monthly change at your current token volume. The quality tier is shown so this is not a blind race to the cheapest option.
Move
→ to
Pick two models to compare.
Model mix
Where your tokens and spend actually go, by model.
Loading…
More lenses
Additional cost intelligence Helix surfaces as your data accumulates.
Waste detector
Tokens burned on failed or retried calls this period, per agent. Loading…
Spend anomalies
Day-over-day spend spikes versus each agent's own 7-day median, to catch a runaway loop early. Loading…
Efficiency score BETA
Output tokens per EXC and tokens per successful outcome, so you can see which agents deliver more per credit.
Cache & dedup SOON
Near-identical prompts within a session that could be cached, and downgrade candidates whose task complexity does not justify a premium model.
The model assurance layer. The certified model your workforce runs on, its in-country failover chain, and every time work moved from one model to another. Configuration lives in Account → Models & Failover; this is the live read-out.
Active configuration
The organisation default. KEYSTONE refuses any model that would leave your jurisdiction.
Loading…
Failover events
Every time the primary model slowed or failed and work moved to a certified in-country backup. An empty list is a healthy list.
Loading…
Certified models
Models proven against your own tasks, cheapest first. Only certified models can run or act as a failover.
Loading…
What your loops found, file by file, at every level the loop declares (for a Game Plan review: licence holder, agreement holder, broker). Every number is counted from the findings each run published, at the time it ran. Click a row to go one level down; at the bottom are the files, each with its report. Save a view to come back to it, and pin it to the CRO lens of the Control Plane.
Findings
Choose a loop.
Repeat issues
The same issue failing on more than one file, at the level chosen. This is where coaching starts.
Saved views
Pinned views appear on the CRO lens of the Control Plane, recomputed each time it opens.
Job posts
Double-click a role to see everything that was posted.
No jobs posted yet
Post your EightX internal roles to start building your AI workforce
Post first role →
Role
Hire Requests
Requester
Agent requested
Company
Received
No hire requests yet — they appear here when employers click Hire on find.html
ECHO
Workforce memory
Every agent has three memory namespaces: episodic (employer-owned), semantic (anonymisable), procedural (agent-owned, travels with passport). Configure your default policy here; override per-agent below. Powered by Mem0 vector store + your local Postgres.
Total Memories
–
across workforce
By Type
–
episodic · semantic · procedural
Storage
–
total bytes used
Monthly Storage Cost
–
EXC · debited daily
Write 1 EXC ($0.01) episodic/semantic · 2 EXC ($0.02) procedural
Read 0.5 EXC ($0.005)
Storage 0.001 EXC/memory/day
List/prune free
Default Policy
Applied to every agent unless overridden
Episodic memory
Conversations, tasks, observations
days
Semantic memory
Derived knowledge, patterns
days
Procedural memory
Skills, agent-owned (forever)
no expiry
I/O event logging
Audit trail · 0.1 EXC/event
days
Share within workforce
Agents suggest patterns to each other
Receive recommendations
2 EXC per accepted suggestion
Compliance mode
Affects retention + portability
Loading…
Per-Agent Namespaces
Loading…
Agent
Status
Episodic
Semantic
Procedural
Bytes
Override
Recent Operations
Loading…
When
Op
Type
Agent
Result
EXC
Cleaner correlation_id audit trails coming post-launch · Cross-agent pattern recommendations land in Phase 2
Delivery Pod
OPERATORStand up a delivery team, run it, then harvest what it learned into the next template version. Visible only to agnt8x and granted delivery operators.
Workspace: …
You: —
Create a delivery pod
Instantiates the nine delivery agents from the delivery-pod template, each with its Passport and role lessons.
template: delivery-pod (latest)
Pods in this workspace
Loading…
Harvest review
Delivery craft lifted at decommission. Promote what generalises into the next template version. Anything classed client is withheld and cannot be promoted.
Compliance
The Control Tower does two jobs. 1. It watches your agents in real time, so their actions do not contravene the regulations and policies you must follow. 2. It recommends the agents and systems you can deploy to evidence and manage compliance across your whole organisation.
Your organisation's regulatory wording, held once and versioned: NCCP sections, RG extracts, your own control wording.
How it is used: a workflow loop setting written
reg:<key> reads the wording by its key, and a compliance control linked to a key judges agents against it. When you change wording you see every loop and agent it reaches before you save, and each loop keeps running on the wording it was proved on (or pauses, if set to) until its owner accepts the new version.Loading…
Your library of regulatory and bespoke policies, grouped by region. Conduct policies are checked live against agent actions; posture policies are attested with evidence. Custom policies you build sit here too.
What to do: pick a policy and click Apply to choose which agents it covers (all, some or none). Double-click any tile to read the full policy, open the official document and see every control, or build your own with + Build your own policy.
Apply a policy to all agents, a selected set, or none. Conduct policies are evaluated live against agent actions; posture policies are attested. Use Manage scope on each policy.
Generate a policy from a regulatory document
Drag in a PDF or text file of a regulation (or paste its text). agnt8x extracts the material controls and creates a new, editable policy tile. Review and adjust the controls after import.
📄
Drop a regulation PDF or text file here, or click to choose
No file selected
Or paste the regulation text:
Build a bespoke policy
Controls — written in plain English. The engine evaluates each agent action against these.
Loading policy library…
The engine that evaluates your agents. It reads each agent action (every email, tool call and recommendation) and judges it against the conduct controls in the policies you have applied, returning pass, flag or breach with the evidence and the cited clause.
What to do: click Run engine now to evaluate recent actions on demand. It also runs automatically on a schedule. Review the active assignments and the recent verdicts below.
Run compliance evaluation
No runs yet · the engine evaluates real agent actions against assigned policies
Automatic evaluation
Next run: —
Scheduled runs are paced so they don't compete with your live agents.
Active assignments
No policies assigned yet. Assign a policy from the Policy Library tab.
Recent evaluations
Run the engine to see live verdicts.
Your audit record. The evidence ledger lists every pass, flag and breach with the evidence and the cited clause. The attestation register holds your organisational (posture) controls, which are signed off with evidence rather than scored from logs.
What to do: review breaches and flags, link an evidence agent to each posture control, set its status and owner, attach an evidence link, and click Export audit report for filing.
Evidence ledger
No evidence yet.
Attestation register · posture controls
APRA CPS 234 / 230, ISO 27001 and other organisational controls are attested with evidence rather than scored per action.
Assign a posture policy (e.g. APRA CPS 234 / 230) to populate the register.
The dashboard for management and the board: overall compliance rate, flags, open breaches, compliance by framework, the 14-day trend and per-agent performance.
What to do: review the charts, click Generate Board report for a plain-English governance briefing written by the Board Reporting Agent, or Export compliance report to save or print a filing.
—
Compliance rate
0
Actions evaluated
0
Flags
0
Open breaches (7d)
Compliance by framework
14-day trend
Per-agent compliance
No agent evaluations yet.
How the agnt8x platform itself aligns with AI and financial-services standards across regions, so you can show partners, customers and regulators where you stand. Conduct frameworks are operationalised live against agent actions; posture frameworks are attested with evidence, with certification on the roadmap where noted.
What to do: use this as your reference when a counterparty asks how agnt8x maps to a given standard. It is informational, there is nothing to configure here.
Loading alignment…
Projects
WORKSPACEGroup the documents, plans and reference material behind a body of work, then grant agents access so they can read it for context.
Loading projects…
Project templates
ADMINCreate reusable project templates preloaded with documents. Anyone creating a project can start from these, and the documents copy in as downloadable originals.
Loading templates…