Technical Documentation · v1.9.6

Nexus Hub — Tech Docs

Architecture, security, integrations, and troubleshooting reference for IT, DevOps, and platform engineering teams evaluating or operating Nexus Hub.

01 Architecture overview

Nexus Hub is a two-tier system:

The frontend never connects directly to OpenAI, Anthropic, or other third-party model providers. All AI/NLP processing happens in our backend, which uses spaCy (open-source) for semantic analysis.

🔒 Stateless design — no work item content is persisted. Work item title, description, and acceptance criteria are processed in-memory per HTTP request and discarded immediately after the response is returned. The backend holds no copy, no cache, no log, and no backup of work item content. The only data persisted is your subscription state (organization ID, project ID, plan tier).

02 System requirements

RequirementValue
Azure DevOpsServices (cloud, dev.azure.com) Required
Azure DevOps Server (on-prem)Not currently supported
BrowserChrome ≥ 100 · Edge ≥ 100 · Firefox ≥ 100 · Safari ≥ 15
NetworkOutbound HTTPS to marketplace.visualstudio.com and *.hf.space
PermissionsOrganization administrator (for installation only)
Work Item typesUser Story · Bug · Feature · Epic · Task · Issue
Process templatesAgile · Scrum · CMMI · Basic (custom processes supported)

03 Permissions & scopes

The extension requests the following Azure DevOps OAuth scopes:

ScopeAccessPurpose
vso.workReadFetch Work Items (titles, descriptions, AC, history) for analysis
vso.projectReadDiscover teams, area paths, iterations for grouping in dashboards
No write scopes requested. The extension cannot modify, delete, or create Work Items — it only reads.

04 Data handling

Nexus Hub follows a strict stateless processing model. The backend never writes work item content to disk and never replicates it to logs or backups.

What we send to the backend (per request, in-memory only)

What we DO NOT send

What we persist

What we DO NOT persist

Backend lifecycle: Each analysis request creates a short-lived in-memory object that is garbage-collected as soon as the HTTP response is flushed. There is no caching layer, no message queue retention, and no persistent log of work item content. See Privacy Policy for the data controller statement.

05 Security model

06 API integration

The backend exposes the following REST endpoints. All endpoints require the X-Nexus-Org-ID and X-Nexus-Project-Id headers and are served over HTTPS.

EndpointMethodPurpose
/api/v1/analisar_batchPOSTBatch INVEST analysis of work items
/api/v1/monte-carloPOSTRun delivery date forecast
/subscription/statusGETCurrent plan tier & trial days remaining
/checkout/create-sessionPOSTCreate Stripe checkout session
/portal/create-sessionPOSTCreate Stripe customer portal session

Generic request contract

All POST endpoints accept application/json and respond with application/json. Request bodies follow the documented schema (full OpenAPI spec available on request); responses include a request_id for correlation.

Generic response shape

{
  "request_id": "req_8e2…",
  "status": "ok",
  "data": { /* endpoint-specific payload */ },
  "meta": {
    "plan": "pro",
    "trial_days_left": 9
  }
}

Backend base URL: https://jonathas3837-ato-agile-intelligence.hf.space

Note: These endpoints are designed for the Nexus Hub extension client. Direct API access for custom integrations (CI pipelines, in-house dashboards) is available on request — contact support for credentials and the OpenAPI spec.
Source code: This document describes the public API contract only. The proprietary backend implementation (NLP pipeline, Monte Carlo engine, scoring weights) is not distributed and is not part of the published extension.

07 Billing & licensing

08 Performance & limits

OperationFree planPro plan
INVEST analysis (per item)~300ms~300ms
Monte Carlo simulations / week1Unlimited
Monte Carlo iteration count10,00010,000
Monte Carlo runtime (typical)~2s~2s
Backlog IQ batch size50 items500 items
API rate limit30 req/min/project120 req/min/project
Backend cold start (HF Spaces)~10s if idle > 30min~10s if idle > 30min

09 Troubleshooting

Extension doesn't load / shows blank screen

  1. Check browser console (F12) for errors.
  2. Verify outbound HTTPS to *.hf.space is not blocked by corporate proxy.
  3. Hard refresh: Ctrl+Shift+R.
  4. Try in incognito mode (rules out browser extension conflicts).

Theme doesn't match Azure DevOps theme

Click the toggle button in the hub header (sun/moon icon) to manually pick Light or Dark. The extension persists the choice in localStorage.

Trial expired earlier than expected

The trial is per project. Each project gets its own 14 days from first open. Check Predictive Analytics → trial banner shows exact days remaining.

"Subscription not active" error after paying

Stripe webhooks may take up to 60s to propagate. Refresh after 1 minute. If persistent, contact support with your Stripe receipt number.

Backend timeout / 504 Gateway Timeout

Backend hibernates on Hugging Face Spaces if idle > 30min. The first request after hibernation triggers a cold start (~10s). Retry the operation.

Monte Carlo forecast says "history too sparse"

The team needs at least 6 weeks of throughput data with non-zero weeks > 25%. Either expand the sample window or wait for more data.

10 Compliance

← User Manual · Back to Support →