Queue and identity
Assign one request ID, status, owner, and history to each analysis.
The Hub coordinates the work. The model reviews the evidence. You decide what to verify next.
AI Hub turns a request into a tracked analysis, gathers the right evidence, coordinates the review, and keeps the result together.
Assign one request ID, status, owner, and history to each analysis.
Collect the main record and apply limits before anything reaches the model.
Route qualified parsers or plugins with explicit budgets and an allowlist.
Check the response shape, record evidence references, gaps, and every workflow event.
The LLM reviews the prepared evidence and proposes a direction; it does not own the queue, permissions, or final engineering decision.
The LLM cannot directly operate tools or publish comments; BLUE controls those boundaries.
Hub splits large evidence into focused reviews, checks each response, and combines the findings. Supported local-tool follow-ups add new receipts within limits.
Normalize the request, collect bounded facts, and attach source references.
Return facts, hypotheses, missing evidence, and a suggested next check.
Validate schema, references, allowed scope, and remaining gaps.
Show a result, or prepare one more explicitly recorded review pass.
When another pass needs evidence, return to step 1. Stop at the configured limits and keep any remaining gaps visible.
Supported typed action → local approval/execution → source-bound receipt → next review. Follow-up is available only when a qualified worker and runtime gate are present; it is not universal autonomous execution.
BLUE reduces avoidable errors by controlling what the model sees and what it is allowed to return.
Ticket text and logs are treated as source material, not as commands to the workflow.
Responses must fit a known shape and point back to valid evidence references.
Tools, rounds, and evidence volume are bounded before execution.
Partial, blocked, or missing evidence is shown instead of silently filled in.
BLUE makes the path from request to review easier to repeat, inspect, and share across the Portal and CLI.
Turn a request into a bounded evidence package with a consistent starting point.
Keep inputs, rounds, evidence gaps, and results together for later review.
Use the Portal or CLI and return to the same analysis record when work continues.
Portal, CLI, or an approved tagged workflow.
Identity, scope, queue, and a recorded path.
Bounded source facts and qualified local receipts.
Findings, hypotheses, gaps, and the next useful check.
Suggestion, evidence boundary, gaps, and history.
A bounded follow-up may add a new local receipt to the next review. Results remain suggestions for engineering verification, not root-cause verdicts.
Ticket scopeTriage normally stays on the requested main record. Linked tickets, video, and Ticket attachments over 10 MB are left out; native files stay on the workstation.
AI Hub turns a request into a tracked workflow: it keeps the request identity, chooses the evidence boundary, sequences the work, and records what happened. It prepares the conditions for review; it does not claim the engineering root cause.
The open source LLM reads the prepared evidence bundle and returns observed facts, hypotheses, gaps, and a suggested next step. It does not own request permissions, browser access, or the comment-writing path.
BLUE can use the previous result to prepare another bounded review. When a qualified local worker produces a source-bound receipt, that receipt may become evidence for a recorded follow-up rather than an untracked autonomous action.
The workflow passes bounded source material to the model, checks the response shape and references, and keeps missing evidence visible. These controls reduce avoidable mistakes, but engineers still validate the direction with tests and additional evidence.
BLUE gives a team a repeatable way to prepare a request, review scoped evidence, keep uncertainty visible, and return to the same history record when work continues.
Enter a Jira, PIMS or Confluence reference.
Main record only. Linked tickets, videos and attachments over 10 MB are skipped.
Use the Portal for bounded text logs. Use the CLI for native Windows files.
or
The Portal stores the upload temporarily to hash it and extract bounded excerpts. The model receives excerpts with source SHA-256 and line ranges; raw staging is not sent to the model.
Text can be sent through the Portal. Native Windows files stay local and use the CLI only when a qualified parser is available.
The local processor makes a hash-bound evidence bundle. BLUE controls scope, sends bounded evidence to the model, requests another pass only when the workflow calls for it, and records gaps before a result is shown.
Tell us what would make your analysis workflow more useful. Your signed-in identity is attached automatically.
Submit logs and triage requests from your workstation.
Run one command in your terminal. BLUE installs for your user, then asks for your access code privately.
curl -fsS https://blue-hostxxxx.tail4c6b73.ts.net/install.sh | bash && export PATH="$HOME/.local/bin:$PATH" && hash -rRequires Python 3.10+ and curl. No sudo, pip or manual package download. Your access code is never part of the command.
Enter your personal access code.
Stay signed in for up to 7 days on this browser.