Plexara
Portal Tour
See the Platform, Concept by Concept
One View of the Entire Platform
The admin dashboard is the operator surface. Live metrics for tool calls, active sessions, system health, and audit activity, everything a platform owner needs to know the stack is healthy without reaching for a terminal. The portal is a single-page application built on a complete REST API; every surface you see here has a typed, authenticated endpoint behind it, so anything the portal does, your integrations, agents, and pipelines can do. Explore the full API.
Real-Time Platform Health
Why it matters
Operators need to know at a glance whether MCP traffic is flowing, whether upstream data sources are reachable, and whether errors are spiking. Grepping logs does not scale.
How it works
The dashboard aggregates audit events, connection health, and session metrics into a single pane. P50/P95/P99 latency, error rates, and top tools refresh live from the same structured audit store that powers forensics.
AI Output as Durable Organizational Artifacts
Every chart, dashboard, and report an AI agent produces is saved with provenance: which tool calls produced it, which datasets it drew from, which session created it. Assets are not ephemeral chat output. They are versioned resources that persist, are searchable, and carry lineage.
Your Personal Workspace
Why it matters
Practitioners produce dozens of artifacts per week during AI sessions. Without a workspace, they live in chat transcripts and die there. The workspace makes those artifacts first-class.
How it works
Every artifact created in a session lands in your personal assets view, indexed by type, session, and timestamp. Full-text search across titles and content. One-click to open, share, or add to a collection.
Interactive Dashboards With Live Data
Why it matters
Static screenshots go stale. PDF reports do not drill down. Stakeholders want to click into numbers and see the underlying slices, and they want it to reflect current data.
How it works
Agents emit JSX components that render as interactive dashboards in the portal. Charts are linked to the same governed queries that produced them, so every view carries lineage back to the source.
Formatted Reports for Stakeholders
Why it matters
Leadership reads reports, not notebooks. The output needs to be a readable artifact (narrative plus tables plus charts) without a manual polish step.
How it works
Agents produce structured HTML with tables, headings, and embedded charts. The portal renders them with consistent brand styling pulled from managed brand resources.
Scalable Vector Graphics
Why it matters
Raster images pixelate in slide decks and reports. Vector graphics stay crisp at any size and integrate cleanly into downstream deliverables.
How it works
When agents need a chart embedded elsewhere, they emit SVG. The portal stores it as an asset and exposes it as a shareable, embeddable resource.
Documentation and Analyses
Why it matters
Technical write-ups, runbooks, and analyses need a format that renders cleanly, diffs well, and carries code blocks without fuss.
How it works
Markdown assets render inline in the portal with syntax highlighting, linked headings, and the same provenance footer as every other asset type.
Tabular Exports for Spreadsheets
Why it matters
Finance, ops, and vendors often want data back in Excel. A governed CSV export beats an ungoverned copy-paste every time.
How it works
Agents emit CSV from query results. The portal stores the CSV as an asset with the originating query attached, so anyone consuming the file can trace it back to source.
Platform-Wide Asset Governance
Why it matters
When AI-generated artifacts become the official record, admins need to audit who produced what, what sources it drew from, and whether anything sensitive leaked into an output.
How it works
The admin asset view indexes every asset across every user. Filter by type, owner, connection, or time. Inspect any asset down to the tool call that produced it.
Full Provenance on Every Artifact
Why it matters
An asset without lineage is a claim you cannot verify. When a number in a dashboard is wrong, the fix starts with "where did this come from."
How it works
Asset detail shows the originating session, the tool calls executed, the datasets queried, the agent persona, and the full version history with revert.
Curate Team Knowledge, Not Chat Transcripts
Collections are titled groupings of assets with descriptions. A quarterly board packet. The weekly pipeline health review. Onboarding materials for new analysts. Curate once, share across the team, and the curation itself becomes durable.
Named Groups With Purpose
Why it matters
Assets scattered across sessions are not a deliverable. A collection turns a loose set of artifacts into a titled, described, shareable package.
How it works
Create a collection, add assets from any session, reorder them into sections, attach a description. The collection appears in team views and can be deep-linked.
Structured Presentation
Why it matters
A deliverable is not a bucket. It has a story. Sections, order, and context matter.
How it works
Collections render as sectioned documents. Each section has a title and description; each asset renders inline. Stakeholders see a polished artifact, not a file list.
Drill-Down Without Leaving Context
Why it matters
When a stakeholder asks "where did this chart come from," they want the answer without a context switch.
How it works
Every asset in a collection is clickable to the asset detail view, full provenance included. Come back to the collection with one click.
Tribal Knowledge as Structured Intelligence
Every correction an analyst makes, every bit of business context shared in a session, every quality observation: captured, reviewed, synthesized, and written back to the metadata catalog. Your organization gets smarter with every conversation.
The Knowledge Control Plane
Why it matters
Knowledge capture is a workflow, not a feature. You need to see what is pending, what has been approved, what has been rejected, and what is flowing downstream.
How it works
The knowledge overview surfaces counts of pending changesets, recent approvals, coverage by domain, and active contributors. Admins route, review, and audit from one view.
The Living Metadata Catalog
Why it matters
A static data dictionary goes stale the day it ships. A living catalog that updates from real usage is the only version that survives contact with reality.
How it works
Approved insights become entity-level annotations: column meanings, business rules, quality flags, ownership. Every AI session reads them. Every AI session can propose updates.
Governed Review Before Anything Writes
Why it matters
Letting AI update the catalog unattended is how trust evaporates. Every change must route through a human.
How it works
Changesets are proposed edits: what is being added, what is being changed, the source session, the confidence score. Reviewers approve, reject, or revise before anything is applied.
Persistent Context Across Sessions
Why it matters
Without persistent memory, every new session starts from zero. The AI forgets what it was taught yesterday. Teams re-explain the same business context on loop.
How it works
Memory stores knowledge, events, entities, relationships, and preferences across sessions. Multi-strategy recall (entity lookup, semantic similarity, lineage traversal) surfaces the right prior context automatically.
Your Slice of What the Platform Knows
Why it matters
Analysts need to see what context the AI will bring into a session on their behalf, and need to trust it is accurate.
How it works
The personal knowledge view shows what is available to your persona: accessible entities, glossary terms, ownership, and the memory items that will inform recall.
Repeatable Workflows, Not Rewritten Instructions
Prompts are typed, parameterized workflows stored at global, team, or personal scope. "Produce the Monday pipeline health report." "Draft a data quality summary for {dataset}." Shared templates turn one-off prompt engineering into durable institutional workflows.
Governed Prompt Library
Why it matters
Ungoverned prompts mean every analyst invents their own, quality drifts, and nobody knows which prompt produced which report.
How it works
Admins publish curated prompts at global scope, teams publish at persona scope, individuals keep personal ones. Typed arguments, descriptions, and usage telemetry track adoption.
One-Click Workflows in Session
Why it matters
Typing a 300-word prompt every time you want the quarterly report is a waste of everyone involved.
How it works
The user prompt surface exposes platform, team, and personal prompts with a run button. Fill the typed arguments, execute, receive a governed result. No prompt engineering required for repeat work.
Security as the Default Operating Mode
Personas map identity provider roles to platform capabilities. Analysts get analytics tools. Executives get curated summaries. ETL services get pipeline access. Default-deny: if you are not explicitly granted access, you get nothing.
Identity-Driven Capability Mapping
Why it matters
Access defined by technical permissions drifts from business reality. Access defined by role (analyst, executive, service account) stays coherent with how the organization actually operates.
How it works
Personas map upstream IdP roles to tool allow-lists, connection allow-lists, and prompt scopes. Fail-closed: unauthorized tools and sources are not even visible to the agent, reducing token waste and blocking access attempts before they reach the data layer.
Every MCP Tool, Typed and Testable
MCP tools are not magic. They are typed RPC endpoints. Inspect the catalog, read the schemas, execute with dynamic parameter forms, and see results inline. The tool surface is the debugger.
Full Tool Catalog With Telemetry
Why it matters
A tool you cannot introspect is a tool you cannot govern. Schema visibility and usage telemetry are prerequisites for safe operation.
How it works
Every MCP tool is listed with its JSON schema, description, and call counts. Filter by connection, by persona allow-list, by error rate.
Execute Any Tool With a Form
Why it matters
Debugging tool behavior by asking an agent to call it is indirect and slow. You want to execute the tool directly with the same inputs and see the exact response.
How it works
The tool explorer generates dynamic forms from each tool schema. Fill in parameters, execute, see the full response including enrichment payload and audit entry.
Your Data Estate as a Single Queryable Surface
Connections are the bridge between Plexara and your data sources: Trino clusters, DataHub instances, S3 backends, other MCPs. Managed at platform scope with persona-based visibility and read-only enforcement where it matters.
Every Upstream System, Centrally Managed
Why it matters
Credentials scattered across agents, services, and config files is a breach waiting to happen. Centralization is a security requirement, not a convenience.
How it works
Connections hold credentials, query limits, read-only flags, and allow-list patterns. Health checks run continuously. Rotation is a config change, not a hunt through source.
MCP Resources Beyond Queries
Why it matters
Not everything is SQL. Brand assets, reference documents, schemas, and managed files need to be discoverable and governable too.
How it works
Resources are first-class MCP primitives: browsable by agents, enriched with metadata, protected by the same persona model as tools.
Discoverable Context for Every Session
Why it matters
Agents need access to reference material (brand guides, schema docs, example queries) without humans copy-pasting it in every time.
How it works
The user resources surface shows what is available to the active persona. Agents discover them automatically through MCP; humans see the same list for transparency.
An Audit Log You Can Actually Query
Grep over log files is not an audit system. Every tool call, every query, every error flows through structured audit middleware with indexed fields for user, tool, timestamp, status, and parameters. Real-time metrics. Searchable events. Drill-down on every interaction.
P50/P95/P99 Latency and Error Rates
Why it matters
Incident response starts with knowing whether the stack is degraded, and for whom. Aggregate metrics are the first 30 seconds of every page.
How it works
The audit overview charts latency percentiles, success rates, top tools, top users, and error trends. Refresh is live. Time ranges are arbitrary.
Every Call, With Full Parameter Detail
Why it matters
When something goes wrong, you need the exact tool call, the exact arguments, the exact response, and the exact session, not a sanitized summary.
How it works
Event search indexes every tool invocation. Click any row for a detail drawer with inputs, outputs, timings, persona, and session linkage. Export the search as a report.
Enterprise Operations, Not Afterthoughts
API keys for service accounts. Structured changelogs for platform updates. The boring plumbing that makes a system operable in a Fortune 500, rather than a demo that dies the first time someone rotates a secret.
Service Accounts, Rotatable and Auditable
Why it matters
Shared credentials are how breaches happen. Per-service keys, scoped to personas, rotatable on a schedule, are how breaches do not happen.
How it works
Keys are minted per service with persona scope, expiration, and usage tracking. Rotation is a one-click operation with a grace window. Revocation is instant.
Structured Record of Platform Evolution
Why it matters
Platform updates that appear without notice break trust with operators. A changelog is a commitment to transparency.
How it works
Every platform-level change (new tool, policy update, schema migration) lands in a structured changelog with scope, affected components, and release notes.
Visibility Into What the Team Is Doing
Activity feeds and shared asset surfaces turn individual work into team-visible output. See what colleagues have produced, consume their published collections, and trace any shared artifact back to its governed origin.
Your Recent Work, Chronologically
Why it matters
Analysts often want to pick up where they left off or find that dashboard they built last Tuesday. A chronological feed beats folder navigation.
How it works
The activity view shows recent assets, sessions, and shared items with time-range filters and quick links to open or re-share.
What the Team Has Published
Why it matters
Good work buried in a personal workspace might as well not exist. Published assets and collections are discoverable by the team, with access governed by persona.
How it works
Shared surface lists everything the active persona can see: collections from peers, broadcast assets, and organizational packages. Full-text search across titles and descriptions.
Consume Peer Output With Full Provenance
Why it matters
Using a peer-produced chart should be one click, and it should carry the same lineage and quality signals as something you produced yourself.
How it works
A shared asset view opens with full provenance, source session, and the ability to fork into a new session for follow-up analysis.
