Rangor

The context layer for AI agents

What your team needs to know —
packaged for the model.

Rangor connects the systems your team already uses — tickets, wikis, repos, files on Google Drive — and merges linked sources into one unified package per task: what's decided, what's contested, what changed. After you ship, Usage Receipts record which claims helped — so the next package ranks smarter for your team.

Get started See the product

Sign up in minutes · connect a source · no credit card

Steward of UCP Open spec at ucpcore.org Fully managed SaaS

15×smaller than raw threads
One UCPfrom every linked source
Usage Receiptsteach ranking what your team cites
PROJ-482 · Sidebar
What matters most

Security changes before the Aug 1 launch

1Incoming webhooks must be signed — required by the security review
2Old API keys stop working Aug 1 — migration guide is linked in the ticket
3Duplicate-request protection was discussed — not planned for this release
3 decisions

What's already been decided

yesWebhook signing is mandatory — agreed Apr 12
yesRetire old API keys on Aug 1 — agreed May 3
noExtra duplicate protection — postponed to a later quarter
2 open disagreements

Teams still don't fully agree

·Security wants signing before launch — some engineers say internal hooks can wait
·Hard Aug 1 deadline for old keys — platform team asked to extend until October
Linked from 6 places

Every fact points to a real document

JiraTicket PROJ-482
WikiSecurity policy SEC-44
GitHubPull request #318

Your AI is working blind

The full story lives across tickets, shared drives, wikis, and threads — but your agent only sees what you remember to paste.

Without Rangor

You hunt. You paste. The model fills the gaps.

  • Hunt across five tabs Jira, Drive, Slack, wiki — nothing wires itself to your agent.
  • Siloed by tool Jira here, Confluence there, PR on Git — three tabs, three pastes, nothing connects itself.
  • Too little or too much Skip a doc and the AI guesses. Dump everything and it drowns in noise.
  • Decisions get buried What was agreed? What's still disputed? Lost in comment thread #47.
  • Same work, every task Every ticket means rebuilding the same brief from scratch.

With Rangor

Open the task. Get the brief.

  • Full picture in seconds Rangor pulls ranked facts from every connected source — no copy-paste.
  • One package across tools Ticket, wiki page, PR, and Drive file — merged by the graph into one ranked brief. ACL on every object.
  • A brief, not a dump Only what matters for this task, compact enough for any agent.
  • Nothing swept under the rug Decisions, disagreements, and where each fact came from — all visible.
  • Where you already work Sidebar, Cursor, ChatGPT, Claude — same package everywhere.

15× smaller than copying whole threads

Every source → one package

Connectors index into the same temporal graph. Open a Jira key, Confluence page, or PR — Rangor follows links (ticket keys, URLs, team-defined edges) and assembles one unified UCP, not separate fetches per system.

Data sources

+ more planned

Your agents

Cursor ChatGPT Claude Sidebar

Unified context — not seven separate tools

Your feature spans Jira, Confluence, GitHub or Gitea, and Drive — but your agent shouldn't need four separate calls. Rangor follows links in the graph — ticket keys, URLs, remote links, manual team links — and assembles one permission-filtered package. Jira is a convenient hub, not a requirement.

  • Jira PROJ-482 ticket + Confluence spec + GitHub PR + Drive policy
  • Confluence PAY:482 wiki page + Gitea PR + linked files — no Jira required
  • Gitea acme/api#16 PR thread + spec page + manual cross-links

Not search. Not RAG. A context layer.

Glean finds documents. RAG finds similar chunks. Rovo stays inside Atlassian. Rangor assembles the state of the task — permission-filtered, provenance-backed, ready for your agent.

Glean · enterprise search

Query → doc links

You ask a question; it returns search results and chat answers. Synthesis stays on you — no task trigger, no structured package for agents.

RAG pipelines

Query → similar chunks

Embeddings retrieve text that looks like your prompt. Stale and current facts rank alike — no decisions, conflicts, or temporal validity.

Atlassian Rovo

Vendor-native AI

AI inside Jira and Confluence. Strong within Atlassian — but not a cross-source context graph with portable output for Cursor or other agents.

Rangor

Task → UCP package

Open any linked ticket, page, or PR — no query required. The graph merges neighbors from every connected system into one UCP: must-know facts, decisions, conflicts, and context_diff — to Cursor, ChatGPT, Claude, or Sidebar.

How Rangor compares to enterprise search, RAG, and Atlassian Rovo
Glean RAG Rovo Rangor
Trigger User query User query Ask in Atlassian Open task / issue
Output Search results, chat Text chunks Chat in Jira/Confluence Structured UCP
Knowledge model Search index Vector store Atlassian data Temporal graph
Cross-source assembly Separate searches Separate chunks Mostly Atlassian-only One UCP from linked sources
What changed? — — — context_diff
Conflicts visible — — — Explicit section
Learns from your team — — — Usage Receipts → warm ranking
Agent delivery API / chat UI Custom pipeline Atlassian UI MCP, Actions, UCP

Four questions search and RAG can't answer

What's true now?

Temporal graph with valid-from / valid-to — invalidated facts don't resurface in top results.

What changed since last visit?

context_diff against your last view of the task — not a fresh search every time.

Why was it decided?

Decisions layer with provenance — not buried in comment thread #47.

What contradicts?

Conflicts from graph invalidations stay explicit instead of being merged away.

Five things only a context layer does

Search finds documents. RAG finds chunks. Rangor assembles the state of the task — and keeps improving it for your team.

01

One unified package

Start from a ticket, wiki page, or PR — Rangor traverses the cross-source graph and pulls linked neighbors into a single UCP. Confluence + self-hosted Git works without Jira. Deterministic links, ACL on every object.

02

Permissions travel with context

ACL from every connected source syncs to every package. You never see a fact from a file or page you can't open — filtered per user, not per workspace.

03

Every claim is verifiable

Schema-valid UCP output: each fact cites a source, each source carries a sha256 hash. Disagreements stay visible instead of being merged away.

04

Context that learns

Usage Receipts record which claims your agent cited. Warm ranking boosts what helped your team — privacy-safe, claim IDs only.

05

Open format, managed platform

Rangor ships UCP — the open Universal Context Package we steward at ucpcore.org. Portable structure, fully managed SaaS at app.rangor.io.

Inside every package

One JSON document. Predictable sections any agent — or human — can consume. MCP carries access; UCP carries understanding.

  • must_know — ranked facts with salience scores
  • decisions — what was decided, when, and status
  • conflicts — contradictions kept explicit
  • context_diff — what changed since your last visit
  • usage_receipts — feedback loop for warm ranking (claim IDs, not text)
  • related_objects — cross-source neighbors merged into one package
  • sources — hashed provenance for every claim
Read the UCP spec

See what changed since your last visit

Search and RAG start fresh every time. Rangor remembers when you last opened a task and highlights only the delta — new decisions, moved deadlines, resolved conflicts.

Glean and typical RAG can't answer: «what changed since I was here Tuesday?»

PROJ-482 Last visit · Tue 2:14 PM

You open the ticket again three days later…

Same engine, different teams

Real workflows — not generic search. Each story is a task brief Rangor assembles from your connected sources.

Engineering

PROJ-482 · webhook security

Before

47 Jira comments, two Confluence pages, a GitHub thread — pasted into Cursor until something sticks.

With Rangor

Ranked brief: signing required, old keys retire Aug 1, one disagreement still open — ~1,200 tokens.

From Jira + GitHub + Confluence

Legal

CASE-104 · vendor MSA review

Before

40-page MSA on Drive. Associate greps for «indemnity» — misses the playbook update from last quarter.

With Rangor

Cap matches playbook v3, CFO exception cited from email thread, two clauses still in redline — sources linked.

From Google Drive + matter tracker + policy wiki

Finance

FIN-Q3 · revenue recognition

Before

Close memo quotes July policy. The October revision lives in a Drive comment nobody re-read.

With Rangor

Policy delta surfaced automatically, audit sign-off linked, spreadsheet exception flagged as a conflict.

From Google Drive + approval threads + policy docs

How it works

Connect once. Every task gets a ranked, permission-filtered package — in Sidebar, Cursor, ChatGPT, Claude, or the Dashboard. Submit Usage Receipts after you ship so ranking learns your team.

01

Connect sources

OAuth or tokens for the systems your team already uses. Indexing runs in the background from the Dashboard.

JiraConfluenceGitHubGoogle DriveGitea

02

Build task context

Cross-source graph + salience ranking — no LLM in the ranking loop (graph, heuristics, Usage Receipts). Structural conflicts come from the temporal graph; optional LLM enrichment on generate can surface disagreements buried in comment prose.

Ranking decides what matters most. LLM enrichment, when enabled, helps with semantic synthesis and prose-only conflicts — not salience order.

03

Use anywhere

Browser Sidebar, Cursor, ChatGPT, Claude, or Dashboard — same package, same guarantees.

04

Submit Usage Receipts

Mark which claims your agent used and whether the outcome worked. Warm ranking boosts what helped — claim IDs only, never your document text.

One context layer, every agent

Same engine underneath — including the learning loop from Usage Receipts. Pick how your team loads task context.

Context beside the ticket

Chrome extension side panel on Jira, GitHub, and more. Ranked must-know facts, decisions, and conflicts without leaving the page.

  • Auto-detects issue from the open tab
  • Submit Usage Receipts after you ship
  • Connects to your Rangor workspace

Proof: measured on real issues

Same token estimator (~4 chars/token). Raw thread = everything you'd paste into the model. UCP = ranked task context with decisions and provenance intact.

microsoft/vscode#519 · 596 comments over ten years
raw thread~18,500 tokens
ucp package~1,200 tokens

15× smaller — with decisions, conflicts, and hashed sources.

More benchmarks

Issue Comments Raw thread UCP
microsoft/vscode#519 first 200 of 596 ~18,500 ~1,200
rust-lang/rust#158622 12 ~4,450 ~1,450
pallets/flask#5961 4 ~800 ~700
pallets/flask#5948 0 ~500 ~330

The win grows with thread size: a decade-long discussion collapses ~15× while keeping decisions, conflicts, and provenance. Reproduce with benchmark_context.py or try live at ucpcore.org/try.

Try the format in browser Sign in to try

Built on open standards — run as managed SaaS

No fake logos. Here's what you can verify today.

Authors of UCP

We steward the open Universal Context Package spec at ucpcore.org — portable format, vendor-neutral protocol.

Reference platform

Rangor is the hosted implementation at app.rangor.io — indexing, graph, ranking, and MCP included.

Permissions first

Every package filtered per user from source ACLs. Fail closed — no data you can't already open.

Privacy-safe learning

Usage Receipts improve ranking with claim IDs only — not your document text.

Reproducible benchmarks

Token savings claims link to public scripts and live demos — try the format yourself.

Beta, clearly labeled

v0.5.0-beta — features and pricing evolve in the open. Enterprise SLAs on request.

Enterprise trust, built in

The first question from every security team: «will the AI see data my user can't access?» Rangor's answer is structural — not a policy page.

Permission-aware by design

Relationship-based access control syncs permissions from every connected source. Context packages are filtered per user before delivery.

Full provenance

Every claim links to hashed sources. Temporal graph tracks valid_from / valid_to — contradictions surface instead of hiding.

Fully managed SaaS

Hosted at app.rangor.io — connect your sources, we run indexing, graph, and MCP. No servers to provision.

Audit & team workspace

Shared index per organization with per-user visibility. Access audit for compliance reviews.

We wrote the standard

Rangor is the reference SaaS for UCP — Universal Context Package. Open spec at ucpcore.org; managed context layer at app.rangor.io. Protocol and product, separated on purpose.

Explore UCP spec

A new category — and a compounding moat

For teams evaluating category creation vs incremental search AI.

Category

Context layer, not search

Search answers a question. Rangor assembles task state — triggered by opening a ticket, not typing a query.

Architecture

Precomputed temporal graph

Knowledge indexed ahead of time — enables context_diff, conflicts, and fast package assembly. Not on-demand RAG per request.

Flywheel

Open format · closed ranking

UCP is open at ucpcore.org. Team-specific warm ranking from Usage Receipts — data moat that doesn't export.

FAQ

Plain answers to what teams ask before they connect their first source.

Is Rangor just RAG?

No. RAG retrieves similar text chunks for a query. Rangor builds the state of a task from a precomputed graph — what's decided, what's disputed, what changed since your last visit — without you writing a prompt.

How is this different from Glean or Copilot in Jira?

Enterprise search answers questions across documents. Copilot summarizes inside one vendor. Rangor connects multiple sources and delivers a structured, permission-filtered package to Cursor, ChatGPT, Claude, or any agent — triggered by the task you opened, not a chat box.

Do you train AI models on our data?

No. Your content is used to build context packages for your workspace. Usage Receipts store claim IDs and outcomes for ranking — not document text.

Can we self-host?

Rangor is fully managed SaaS at app.rangor.io. Enterprise plans offer dedicated tenant isolation — contact us for deployment options.

What's included in the free plan?

Full UCP format, Sidebar, Cursor MCP, ChatGPT & Claude Actions, one data connector (GitHub public or one Jira project), 50 packages per month, and polling sync. Upgrade to Team for all connectors, shared workspace, and unlimited packages.

Will the AI see documents I can't access?

No. ACLs from Jira, Drive, GitHub, and other sources sync into every package. Context is filtered per user before delivery — fail closed.

What sources do you support?

Jira, Confluence, GitHub, Gitea, Bitbucket Server, and Google Drive today — with more connectors on the roadmap. Connect in the Dashboard with OAuth or tokens.

Does Rangor use LLMs?

Ranking — what surfaces first in must_know — uses the temporal graph, heuristics, and Usage Receipts. No LLM in that loop. Optional LLM enrichment on generate adds semantic synthesis and can extract conflicts or decisions stated only in comment prose; graph-based conflicts and cross-source merge work without it. Provenance is preserved either way.

Is annual billing available?

Yes — Team and Business include annual prepay at 20% off per seat ($240 and $432 per seat per year — $20/mo and $36/mo equivalents). Use the Annual toggle above; Enterprise is custom annual contracts via sales.

Is each connector a separate context call?

No. Every source indexes into the same graph. When you open any linked item — Jira key, Confluence page, GitHub or Gitea PR — Rangor assembles neighbors from other systems into one UCP. Links come from ticket keys, URLs in text, Jira remote links, and manual team links you confirm in the Dashboard — not from the model guessing.

Do I need Jira as the hub?

No. Jira is a common anchor, but you can start from Confluence, GitHub, Gitea, Bitbucket, or Drive. If the graph connects a wiki page to a self-hosted PR and a policy file, one context load returns all three — permission-filtered and ranked together.

How does Rangor learn over time?

When your agent uses a package, you submit a Usage Receipt — which claim IDs were cited, dismissed, and whether the outcome worked. Warm ranking uses those signals inside your workspace only. No document text is stored in receipts; the next package for similar tasks ranks higher on what actually helped your team.

How do I connect ChatGPT or Claude?

Open the Dashboard → Agent connectors. For ChatGPT: import the OpenAPI schema into a Custom GPT and complete the OAuth wizard. For Claude: import the same schema into a Project and authenticate with a token. Cursor and other IDEs use MCP from the same page.

What is UCP?

Universal Context Package — the open JSON format for task context we publish at ucpcore.org. Rangor is the reference SaaS; any agent can consume UCP.

Pricing

Sign up in minutes — no credit card. Every tier gets the full UCP format; limits are on scale, sources, and team features.

Solo developer

Free

$0

  • Full UCP — decisions, conflicts, provenance
  • Cursor MCP + Sidebar + ChatGPT / Claude Actions
  • 1 connector (GitHub public or 1 Jira project)
  • 50 packages / month
  • Polling sync (~15 min)
  • Usage Receipts
Get started

Teams · 3–50 seats

Team

$25 / seat / mo

  • Shared workspace & invites
  • All connectors (Jira, Drive, GitHub…)
  • Unlimited packages
  • Webhooks + incremental sync
  • Warm ranking & team learning loop
  • Permission-aware access control
Get started

Security & compliance

Business

$45 / seat / mo

  • Everything in Team
  • SSO / SAML
  • Audit log export
  • Admin roles & access review
  • Priority support
Get started

Regulated & large orgs

Enterprise

Custom

  • Everything in Business
  • Dedicated tenant / isolated infra
  • Custom SLA & DPA
  • Security review & compliance pack
  • Multi-org & volume billing
Contact sales

Beta pricing (v0.5.0) — limits and tiers may change. Team and Business: monthly or annual prepay (−20%). Enterprise: custom annual contracts. Business SSO on request.

Load your first task context

Create a workspace, connect your sources — then wire Cursor, ChatGPT, or Claude from the Dashboard.

Get started