Skip to main content

Pyrana Cortex v1.1 — Standalone UI User Guide

This guide covers every page of the Pyrana Cortex standalone UI. Each section describes what the page does, explains every column and field, and includes a screenshot of the live system.


Table of Contents

  1. Dashboard
  2. CxUs (Context Units)
  3. Sets
  4. Relationships
  5. Agents
  6. Extract
  7. Query
  8. Credentials
  9. Sources
  10. Test Prompts
  11. Admin

1. Dashboard

URL: /dashboard

The Dashboard is the Claims Evidence page. It tracks quantitative assertions made in the CORTEX V2 academic paper and patent against live system measurements, giving you a real-time view of whether the system is performing as specified.

Summary Bar

At the top of the page, a row of pill buttons shows how many claims fall into each status:

Badge colourMeaning
BlueGood — within acceptable range
GreenExcellent — exceeds target
YellowWarning — approaching threshold
RedAlert — threshold breached
Gray / PendingNo measurement data yet

Claim Cards

Below the summary bar, cards are displayed in a responsive grid. Each card shows:

FieldDescription
Category labelOne of: Latency, Retrieval Quality, Memory, Ambient, Scale
Source badgePaper and/or Patent — which document the claim comes from
Claim titleThe specific assertion being tracked
Status badgeColour-coded to the summary bar above
Claimed valueWhat the paper or patent asserted
Latest valueWhat the live system measured most recently
DeltaPercentage difference between claimed and measured
Metric unitUnit of measurement, e.g. ms, %

Filtering Claims

Category tabs (Any / Latency / Retrieval Quality / Memory / Ambient / Scale) — click to filter the card grid. The active tab is underlined. The count next to each tab shows how many claims are in that category.

Source pills (Both / Paper / Patent) — restrict the view to claims from a specific source document. The count in each pill updates to reflect the active category filter.

Both filters can be combined — for example, "Latency" and "Paper" shows only latency claims sourced from the paper.

Claim Detail

Click any card to open the detail page at /dashboard/{claim_id}. The detail page shows the full claim text, measurement history, and baseline configuration.

Alert Thresholds

Hover over any card to reveal a gear icon in the top-right corner. Click it to open the Threshold Modal.

FieldDescription
Warning thresholdValue at which the card turns yellow. Leave blank to use ±10% of the target.
Alert thresholdValue at which the card turns red. Leave blank to use ±10% of the target.

The direction of comparison ("≤ target" or "≥ target") is fixed by the claim type and cannot be changed here. Click Save to persist. The page refreshes automatically.


2. CxUs (Context Units)

URL: /cxus

CxUs are the atomic units of knowledge in the system. Each CxU contains a single, verifiable claim with optional supporting evidence and metadata. This page is the central list of all CxUs in the knowledge base.

Status Filter Tabs

The tabs at the top of the list filter by lifecycle state:

TabShows
AllEvery CxU regardless of status
ActiveCxUs included in query results
DraftCxUs pending review — not yet in queries
InactiveCxUs that have been deactivated
VersionedCxUs superseded by a newer version

List Columns

ColumnDescription
CIDContent-addressable identifier (truncated). Click to open the CxU detail page.
ClaimThe claim text, truncated to fit. Full text is on the detail page.
StatusLifecycle state badge — Active (green), Draft (yellow), Inactive (gray), Versioned (blue).
ClassificationGovernance tier — see table below.
VersionCurrent version number shown as v1, v2, etc.
SetsSet memberships. Up to 2 are shown; "…" indicates more exist.

Classification Values

ClassificationMeaning
ABSOLUTENon-negotiable fact. Highest governance level — requires approval to modify.
REGULATEDRequires an approval workflow before activation.
CONTEXTUALStandard classification. No approval required.
TRIBALOrganisation-specific knowledge not derived from documented sources.
DERIVEDInferred or computed from other CxUs.

Creating a CxU

Click New CXU (top right) to open the creation form.

FieldRequiredDescription
Claim textYesState the claim clearly and concisely.
ClassificationNoDefaults to CONTEXTUAL.
Supporting contextNoOptional verbatim quotes or evidence that back the claim.
Entity IDNoScope the CxU to a specific entity (e.g. a product, regulation, or client).
Initial statusNoStart as DRAFT (default) or immediately ACTIVE.

Click Create CXU to submit. On success you are taken to the new CxU's detail page.

CxU Detail Page

URL: /cxus/{cid}

The detail page shows the full CxU including:

  • Full CID
  • Complete claim text
  • Supporting context — the verbatim quote(s) from the source document that back the claim
  • Classification and status badges
  • Activate / Deactivate button
  • History button

Activating a DRAFT CxU — click Activate. The status changes to ACTIVE and the CxU becomes available in query results.

Deactivating an ACTIVE CxU — click Deactivate. The status changes to INACTIVE and the CxU is excluded from future queries.

Version conflicts — if another user has modified the CxU since you loaded the page, a conflict dialog appears. Reload the page before retrying.

Version history — click History to view the change log at /cxus/{cid}/history. Each entry shows the version number, date/time, status at that point, who made the change, and any recorded reason.


3. Sets

URL: /sets

Sets are named groupings of CxUs. They let you scope queries, organise knowledge by topic or domain, and designate certain CxUs as always-included (Rank 0).

Set List Columns

ColumnDescription
NameSet name (clickable link) with optional description below.
CXUsTotal number of CxUs currently in this set.
Rank0Number of always-included (Rank 0) CxUs in this set. "—" means none.
CreatedDate the set was created.
Last ScanDate of the most recent automated folder scan that targeted this set (see Sources). "—" means no scan has run or no folder is assigned.

Creating a Set

Click New Set (top right). Enter the set name and an optional description, then submit.

Set Detail Page

URL: /sets/{name}

The detail page lists all CxUs that belong to the set. A search bar at the top of the member list lets you filter by CID or claim text.

ColumnDescription
CIDTruncated content-addressable identifier.
ClaimClaim text, truncated.
StatusLifecycle state badge.
Rank0Star icon — filled (★) means always-included; outline (☆) means standard member.

Adding a CxU — enter the full CID into the input field and click Add.

Promoting to Rank 0 — click the outline star (☆) next to any member. It becomes filled (★). Rank 0 CxUs are always returned in query results for this set, regardless of relevance score.

Demoting from Rank 0 — click the filled star (★) to remove Rank 0 status. The CxU remains in the set but is no longer always-included.

Removing a CxU — click the trash icon next to the member. The CxU itself is not deleted — it is only removed from this set's membership.


4. Relationships

URL: /relationships

The Relationships page manages inferred connections between CxUs. Suggestions are generated automatically during document extraction based on co-occurrence and semantic similarity.

Pending Suggestions Table

When suggestions exist, the table shows:

ColumnDescription
SourceSource CxU CID (truncated).
RelationshipSuggested relationship type, e.g. "supports", "contradicts".
TargetTarget CxU CID (truncated).
ConfidenceModel confidence as a percentage (0–100%).
ActionsApprove (green ✓) or Reject (red ✗) buttons.

Approving — click ✓ to accept. The row is removed and the relationship is stored for use by graph retrieval strategies.

Rejecting — click ✗ to dismiss. The row is removed and the suggestion is not stored.

Empty State

When there are no pending suggestions, the page shows "No pending suggestions — Relationship suggestions appear here after document extraction." New suggestions are generated each time documents are extracted.


5. Agents

URL: /agents

The Agents page shows the live status of CORTEX agent processes and any knowledge gaps they have detected.

Agent Cards

Each CORTEX agent is displayed as a card showing:

FieldDescription
NameAgent identifier with a coloured status dot.
StatusCurrent health state — Registered, Active, Healthy, or Degraded.
TypeAgent type identifier, e.g. RETRIEVAL_AGENT, LEARNING_AGENT, AMBIENT_AGENT, ORCHESTRATING_AGENT.

Status dot colours:

ColourMeaning
GreenActive or Healthy — operating normally.
BlueRegistered — started but not yet processing.
YellowDegraded — running but with reduced capability.
RedError or unknown state.

The four agents and their roles:

AgentRole
Retrieval AgentHandles query execution, Q-value reranking, and cross-encoder reranking across the three retrieval strategies.
Learning AgentManages TEMPR memory and Q-value feedback loops — learns from query outcomes over time.
Ambient AgentRuns proactive knowledge gap detection in the background based on recent query patterns.
Orchestrating AgentCoordinates agent communication and task routing via the message bus.

Knowledge Gaps

If the Ambient Agent has detected gaps, they appear below the agent cards with a yellow left border. Each gap describes an area where the knowledge base may be incomplete.

Refreshing

The page auto-refreshes every 30 seconds. Click Refresh to fetch the latest status immediately. The last-refreshed timestamp appears below the button.


6. Extract

URL: /extract

The Extract page lets you upload documents and extract CxUs from them automatically using the LLM extraction pipeline. Two modes are available via tabs.

Single File Tab

Use this tab to extract CxUs from one document at a time.

Supported formats: .pdf, .txt, .md, .docx, .doc, .csv

FieldDescription
Drop zoneDrag and drop a file here, or click to open a file picker.
Entity IDOptional. Scope the extracted CxUs to a specific entity (e.g. a client name, product, or regulation identifier).
Upload & Extract buttonSubmits the file and starts extraction.

Extraction phases:

PhaseWhat you see
UploadingProgress indicator while the file transfers.
ConvertingThe document is being converted to text.
Extracting"extracting · N CXUs found" — the count updates as lenses complete.
StoringCxUs are being written to the knowledge base.
DoneFinal results shown — count of stored CxUs with quality grading scores if enabled.
ErrorError message in red with the failure reason.

Results display — on completion the page shows:

  • Quality Grading (if enabled): key point coverage (how many of the document's key points were captured) and document fidelity score (how well a document reconstructed from the CxUs matches the original).
  • Extracted CxUs: list of all stored units, each showing the claim text, classification, and status.

Click Extract another to reset the form.

Batch / ZIP Tab

Use this tab to process multiple documents in one operation.

Supported formats inside the ZIP: .pdf, .txt, .md, .docx, .doc, .csv

FieldDescription
Drop zoneDrag and drop a ZIP file, or click to open a file picker. Supported file types are listed inside the zone.
Upload ZIP & Extract All buttonSubmits the archive and starts extraction for every supported file inside it.

Progress display — while extraction runs, a file list appears with per-file status:

StatusColourMeaning
queuedGrayWaiting to start.
processingYellowCurrently being extracted.
completedGreenExtraction finished successfully.
failedRedExtraction failed for this file.

A summary line tracks overall progress: "{complete}/{total} complete · {failed} failed".

Click Upload another ZIP to reset after completion.

Timeout: Single-file extraction times out after approximately 3 minutes. Batch extraction times out after approximately 6 minutes. Large documents or archives with many files may approach these limits.


7. Query

URL: /query

The Query page lets you run ad-hoc semantic searches against the knowledge base to find relevant CxUs.

Search Form

FieldRequiredDescription
QueryYesNatural language question or keyword phrase. Example: "What are the safety requirements for..."
Entity IDNoRestrict results to CxUs scoped to a specific entity. Leave blank to search all entities.
Max resultsNoNumber of results to return — 1 to 20, default 10.

Click Search to run. The button shows "Searching…" while the request is in flight.

Results

Results are listed in order of relevance. Each result card shows:

FieldDescription
Claim textThe CxU claim — bold.
ScoreRelevance percentage — how closely this CxU matches the query. Green ≥ 70%, yellow 40–69%, gray below 40%.
ClassificationGovernance classification badge.
StatusLifecycle state badge.
SetsSet memberships — up to 2 shown.
StrategyWhich retrieval strategy returned this result: "template" (pre-built Cypher pattern), "edge-filter" (relationship-filtered Cypher), or "text2cypher" (free-form NLP-to-Cypher). Shown as "via {strategy}".
Rank 0Badge shown if this CxU is a Rank 0 member of one of the searched sets — meaning it was always-included regardless of score.

Empty Results

If no CxUs match, the page prompts you to try different terms, remove the Entity ID filter, or check that relevant CxUs are in ACTIVE status.


8. Credentials

URL: /credentials

The Credentials page manages API keys used by external systems to authenticate with the A2A (Agent-to-Agent) endpoints.

Credential List Columns

ColumnDescription
NameCredential name (clickable).
PrefixFirst few characters of the key followed by "…" — used to identify which key is which without exposing it.
StatusActive (usable) or Revoked (permanently disabled).
Rate limitRequests per minute and requests per day, shown as "N/min · N/day".
Last usedDate the key was last used to make a request, or "Never".
ActionsTrash icon (active credentials only) — click to revoke.

Creating a Credential

Click New Key (top right) to open the creation form.

FieldRequiredDefaultDescription
NameYesA descriptive label for this key, e.g. "prod-integration-1".
DescriptionNoOptional notes about what this key is used for.
Rate limit / minNo60Maximum API calls per minute.
Rate limit / dayNo10,000Maximum API calls per day.

After clicking Create, a modal displays the full API key once only. Copy it before closing — the full key cannot be retrieved again. The modal requires you to click I have copied the key before it can be dismissed.

Important: If the key is lost, you must revoke it and create a new one. There is no key recovery mechanism.

Revoking a Credential

Click the trash icon on any active credential in the list, or click Revoke credential on the detail page. A confirmation dialog warns that the action is permanent. Revoked credentials immediately deny access to any system using that key.

Credential Detail Page

URL: /credentials/{id}

The detail page shows all credential metadata plus an inline editor for rate limits.

Editing rate limits — click Edit next to the rate limit display. Enter new values and click Save, or click Cancel to discard.


9. Sources

URL: /sources

The Sources page manages server-side folder monitoring. When a folder is registered, the system periodically scans it for new documents and automatically extracts CxUs, keeping the knowledge base in sync with the file system.

Monitored Folders Table

When folders are registered, the table shows:

ColumnDescription
PathAbsolute server-side path to the monitored folder (truncated).
SetThe set into which extracted CxUs from this folder are stored.
CXUsNumber of CxUs extracted from documents in this folder.
Last ScanTimestamp of the most recent scan. If the last scan encountered an error, the error message is shown here. "Never" means no scan has run yet.
IntervalConfigured scan frequency in minutes.
StatusHealthy (green), Overdue (yellow), or Error (red) — see below.
ActionsScan Now button and Delete (trash) button.

Status logic:

StatusWhen it shows
ErrorThe most recent scan ended with an error.
OverdueNo scan has run yet, or the last scan completed more than 2× the interval ago.
HealthyLast scan completed within the expected window.

Registering a Folder

Click Register Folder (top right) to expand the registration form.

FieldRequiredDescription
Server pathYesAbsolute path on the server running the Context Engine API — for example, /data/documents/contracts. This is a server-side path, not a path on your local machine.
Target setYesThe existing set to add extracted CxUs to.
Scan interval (minutes)NoHow often to scan — between 5 and 10,080 minutes.

Click Register to save. Click Cancel to close without saving.

Empty State

When no folders are registered, the page shows "No folders registered — Register a server-side folder to begin automatic CXU extraction." This is the normal state for a fresh installation.

Manual Scan

Click the Scan Now button next to any registered folder to trigger an immediate scan outside the normal schedule. The icon spins during the scan. After the scan completes the page refreshes to show the updated timestamp and any errors.

Removing a Folder

Click the trash icon and confirm. The folder is removed from monitoring. Existing CxUs extracted from it remain in the set — they are not deleted.


10. Test Prompts

URL: /admin/test-prompts

The Test Prompts page has two tabs for evaluating query quality and monitoring CORTEX skill health.

Tab 1: Test Prompts

Use this tab to run test queries against the live knowledge base and inspect which CxUs are returned, at what score, and via which retrieval strategy.

FieldDescription
Load a presetDropdown of pre-configured test queries. Selecting one auto-populates the Query field and any associated set filters.
QueryNatural language query to run. Required.
Filter by setPill buttons for each available set. Click a pill to include only that set in the search. Leave all unselected to search across all sets.

Click Run Query. Results are capped at 15.

Results display:

A summary line shows: "{total} results · {rank0} rank0 · searched: {sets} · {time}ms"

Results are split into two groups:

  • Always-included (Rank 0) — CxUs returned unconditionally because they are marked Rank 0 in one of the searched sets.
  • Query Results — remaining results ordered by relevance score.

Each result card shows:

  • Claim text — click the chevron to expand and reveal the full CID
  • Score badge — green ≥ 70%, yellow 40–69%, gray below 40%
  • Rank 0 badge (if applicable)
  • Set badges
  • Strategy tag — which of the three retrieval strategies returned this result
  • Classification badge

Tab 2: Skill Health

This tab shows the health status of all CORTEX agent skills — the learned behaviours that govern how agents retrieve and rank knowledge. Skills are created at runtime by CORTEX agents and evaluated using the /skill-eval CLI command.

When skills exist, the table shows:

ColumnDescription
SkillSkill name (bold) with category below.
AgentThe agent type that owns this skill.
StateLifecycle state — draft, active, evaluated, tuned, or archived.
HealthHealth flag — healthy (green), outgrowth_candidate (yellow), regression (red), not_evaluated (gray).
Pass %Percentage of evaluation test cases this skill currently passes. "—" if not yet evaluated.
Delta ppChange in pass rate since the previous evaluation, in percentage points. Green = improvement, red = regression.
Last evalDate of the most recent evaluation run.

Health flag meanings:

FlagMeaning
healthySkill is performing well against its test cases.
outgrowth_candidateSkill may be ready to be promoted or expanded.
regressionSkill performance has declined — review recommended.
not_evaluatedSkill has never been evaluated.

Lifecycle states:

StateColour
draftGray
activeBlue
evaluatedIndigo
tunedGreen
archivedGray, strikethrough

11. Admin

User Management

URL: /admin/users

The User Management page lists all users who have authenticated with the system. Users are created automatically on first login — they do not need to be pre-registered.

When users exist, the table shows:

ColumnDescription
UserDisplay name (or email) as a clickable link. Email shown below the name.
RolesAssigned roles shown as badges. "No roles" if none have been assigned yet.
StatusActive or Inactive.
Last accessDate of the most recent login, or "—" if the user has never logged in after account creation.

User Detail Page

URL: /admin/users/{id}

Click any user name to open the detail page. All available roles are shown with their current assignment status.

RoleDescription
context_adminFull access to knowledge base management — CxUs, sets, extraction, sources, and relationships.
company_adminOrganisational administrator — can manage users and credentials.
developer_supportRead-only access for debugging and support purposes.

Assigning a role — click Assign next to an unassigned role.

Revoking a role — click Revoke (shown in red) next to an assigned role. If the user is the last company_admin, revocation is blocked with the message: "Cannot revoke: this is the last Company Admin. Assign another Company Admin first."


The left sidebar is present on every page and provides access to all sections:

ItemURLPurpose
Dashboard/dashboardClaims evidence tracking
CXUs/cxusBrowse and manage Context Units
Sets/setsManage named CxU groupings
Relationships/relationshipsReview and approve inferred CxU connections
Agents/agentsMonitor CORTEX agent health
Extract/extractUpload documents and extract CxUs
Query/querySearch the knowledge base
Credentials/credentialsManage A2A API keys
Sources/sourcesConfigure automated folder monitoring
Test Prompts/admin/test-promptsRun test queries and view skill health
Admin/admin/usersManage users and roles