Confluence AI vs openKMS¶
Comparative research on Atlassian Confluence with Atlassian Intelligence and Rovo, versus openKMS as a self-hosted knowledge platform. Sources: Atlassian product and support pages (2025–2026), openKMS repository docs (as of 2026-06).
Related: Goals (vision) · Architecture · Functionalities · RAGFlow vs openKMS · Operational Knowledge Fitness
Executive summary¶
| Confluence + AI (Rovo) | openKMS | |
|---|---|---|
| Primary identity | Team wiki and collaboration hub inside the Atlassian Cloud stack | Open, self-hosted knowledge management system (documents, articles, wiki, KB, ontology) |
| AI strategy | Teamwork Graph + Rovo (search, chat, agents, studio) embedded where teams already work | Dedicated RAG (KB + qa-agent), project agents (Deep Agents workspaces), optional external openkms-skill |
| Best when | You are standardized on Atlassian Cloud and want AI on pages, Jira links, and ~50 SaaS connectors with minimal ops | You need governed corpora, policy lifecycle, parse-heavy documents, and control of data residency on your stack |
| Deployment | SaaS (Cloud); Data Center can sync to Cloud for AI via connectors | Docker / host; PostgreSQL, MinIO, optional Neo4j, separate VLM and qa-agent |
Confluence AI optimizes collaboration velocity inside Atlassian’s graph of work. openKMS optimizes operational knowledge fitness (OKF dimension) across mixed content types with explicit ACL, lifecycle, and evaluation.
What “Confluence AI” means (2025–2026)¶
Atlassian bundles several layers; names overlap in marketing.
1. Atlassian Intelligence (in-product)¶
Embedded generative AI inside Confluence Cloud (and other cloud apps), often surfaced through Rovo:
- Create with Rovo — draft pages, live docs, whiteboards, or databases from natural-language prompts and links.
- Edit in place — tone, grammar, simplify, translate; undo preserves original.
- Summarize — pages and comment threads; audio summary / read-aloud.
- Definitions — inline explanations of team-specific terms (uses wiki context).
- Jira bridge — turn highlights or scanned page content into Jira work items.
Docs: Write or edit content using Rovo.
2. Rovo (cross-product AI teammate)¶
Rovo spans Confluence, Jira, JSM, and connected third-party apps:
| Capability | What it does in practice |
|---|---|
| Rovo Search | Enterprise search across Atlassian + connectors (Slack, Google Drive, etc.); positioned as often unmetered vs credit-consuming chat |
| Rovo Chat | Conversational Q&A and iteration in context |
| Rovo Agents | Task-oriented agents (e.g. meeting insights, brainstorm facilitator); third-party MCP agents (Figma, GitHub, …) after admin enablement |
| Rovo Studio | Low/no-code builder for custom agents and automations |
| Remix (Confluence) | Transform page content into charts, infographics, diagrams; partner flows to Lovable, Replit, Gamma via MCP |
Context is anchored in the Teamwork Graph (work items, people, links across Atlassian and indexed connectors)—not merely raw page text.
3. Atlassian Rovo MCP Server (outbound)¶
A cloud MCP gateway lets external AI clients (Claude, IDEs, etc.) search and fetch Jira/Confluence (and Compass) with the user’s OAuth permissions. Admins control allowed domains; the server acts as a proxy and does not cache page content long-term per Atlassian’s description.
Docs: Use Atlassian Rovo MCP Server.
4. Licensing and rollout¶
- Plans: Rovo rolls out on Confluence Cloud Standard, Premium, Enterprise (phased; Premium/Enterprise earlier).
- Full feature set: Marketing still emphasizes Premium or Enterprise for the complete Rovo experience (search, chat, agents, studio).
- Org control: Admins can deactivate AI; deactivation limits AI-powered chat/agents while some Rovo apps may remain.
- Credits (2026): Organization-pooled Rovo credits per user tier (e.g. 25 / 70 / 150 per user per month on Standard / Premium / Enterprise—verify on Confluence pricing). Rovo Chat and agent requests consume credits (commonly cited ~10 each); Deep Research higher (~100). Overage billing policies have been described as delayed with admin opt-in—confirm current terms before capacity planning.
- Data Center: Rovo Data Center connectors sync DC content to Cloud so AI can run without full migration (infrastructure stays on-prem; AI processing in Atlassian Cloud).
Positioning vs openKMS¶
Confluence + AI¶
Typical journey: teams already write specs and runbooks in spaces → use Create / summarize / Rovo Chat for speed → Rovo Search finds answers across Atlassian + connected SaaS → agents automate meetings, brainstorming, or third-party actions → Jira carries execution.
Strengths: low friction for wiki-native teams, unified search across the toolstack, mature collaboration (comments, permissions, templates, analytics), and vendor-managed AI safety narrative (Responsible Technology Principles, admin toggles).
Limits for regulated or document-heavy KM:
- Page-centric model — PDF corpora and scanned policies are not the same as openKMS’s parse → Markdown → version pipeline.
- Lifecycle — Confluence has history and labels; openKMS emphasizes series_id, effective dates, supersedes relationships, and current-for-RAG semantics for policy domains.
- RAG transparency — Rovo cites sources in product flows; openKMS exposes chunk_index, retrieval_mode, and retrieval_debug on KB search/Q&A for operator-facing audit.
- Vendor lock-in — Cloud, credits, and graph live in Atlassian’s boundary unless you invest in DC + sync.
openKMS¶
Typical journey: content lands in document / article / wiki channels → optional KB index and qa-agent Q&A → knowledge map and glossary for navigation → resource ACL and evaluations for governance and quality.
See index and Goals for the user/org dual mandate.
Architecture contrast¶
flowchart TB
subgraph ATL [Atlassian Cloud — simplified]
CONF[Confluence Cloud]
JIRA[Jira / JSM]
ROVO[Rovo services]
TTG[Teamwork Graph]
MCP_OUT[Atlassian Rovo MCP Server]
CONF --> ROVO
JIRA --> ROVO
ROVO --> TTG
MCP_OUT --> ROVO
SAAS[Third-party connectors]
SAAS --> ROVO
end
subgraph OK [openKMS — self-hosted]
SPA[React SPA]
API[FastAPI]
PG[(PostgreSQL + pgvector)]
S3[(MinIO / S3)]
WK[Worker + openkms-cli]
VLM[VLM server]
QA[qa-agent]
SPA --> API
API --> PG
API --> S3
API --> WK
WK --> VLM
API --> QA
QA --> API
end| Dimension | Confluence + Rovo | openKMS |
|---|---|---|
| Knowledge unit | Page, live doc, whiteboard, database | Document, article, wiki page, KB chunk, FAQ |
| Search index | Atlassian-managed (Rovo Search + graph) | pgvector + global search + optional wiki semantic index |
| AI runtime | Atlassian-hosted models / routing | Your chosen LLM providers via api_models; qa-agent LangGraph |
| Cross-app context | Teamwork Graph + ~50 connectors | Connectors partial; ontology datasets; manual linking |
| External AI access | Atlassian Rovo MCP Server (OAuth) | Personal API keys, openkms-skill, REST API |
| Data residency | Atlassian Cloud regions | Your infrastructure |
Feature comparison¶
Authoring and wiki¶
| Capability | Confluence + AI | openKMS |
|---|---|---|
| Wiki spaces | Spaces (mature templates, analytics) | Wiki spaces (vault import, wikilinks, graph view) |
| AI drafting | Create with Rovo, inline edit, Remix visuals | No inline drafting in the editor; project agents can draft and upsert wiki pages via openkms-skill (with permission) |
| Whiteboards / live docs | Native | Not equivalent (wiki + articles) |
| Bulk import | Marketplace / migrations | Vault zip/folder import; document upload pipeline |
Documents and parsing¶
| Capability | Confluence + AI | openKMS |
|---|---|---|
| PDF / scan-heavy corpora | Attachments; AI reads page content; not a dedicated OCR pipeline | PaddleOCR-VL pipeline, editable Markdown, versions |
| Spreadsheet / mind map | Limited | XLSX, XMIND first-class paths |
| Channel / folder governance | Space permissions | Document channels + resource ACL inheritance |
RAG and Q&A¶
| Capability | Confluence + AI | openKMS |
|---|---|---|
| Grounded Q&A | Rovo Chat / Search over graph + connectors | KB hybrid search + qa-agent (/ask, streamed threads) |
| Chunk-level provenance | Product-dependent; less operator-oriented | chunk_index, retrieval_debug, chunk edit + re-embed |
| FAQs as first-class objects | Not emphasized | KB FAQs + LLM generate + batch save |
| Historical / superseded content | Page history; less RAG-specific | include_historical_documents, lifecycle, is_current_for_rag |
| Quality measurement | Admin analytics; less built-in eval harness | Evaluation runs (search_retrieval, qa_answer, wiki_content_coverage) |
Agents and automation¶
| Capability | Confluence + AI | openKMS |
|---|---|---|
| In-app agents | Rovo Agents, Studio, MCP partner agents, Remix | Project agents, KB Q&A, map HTML designer, eval assist |
| Visual agent builder | Rovo Studio | None (code-first LangGraph) |
| Jira / ITSM actions | Strong native | Via ontology/API; not Jira-native |
| MCP | Inbound (third-party tools) + outbound (Atlassian MCP server) | External skill; no Atlassian-style MCP server product |
Governance and security¶
| Capability | Confluence + AI | openKMS |
|---|---|---|
| Permissions | Space/page restrictions; Guard (add-on) for advanced | Operation RBAC + resource ACL (Security design) |
| Admin ≠ read-all data | Confluence admin models differ by tier | Admin does not bypass data ACL for read/write |
| AI opt-out | Org-level AI deactivate | Self-hosted: control model endpoints and keys |
| Audit | Atlassian cloud audit logs | Your logging; API keys revocable in-app |
Integration¶
| Capability | Confluence + AI | openKMS |
|---|---|---|
| Ingest from Confluence | N/A (you are inside Confluence) | Potential connector target (development plan—sync jobs not shipped) |
| Export to RAG tools | Rovo MCP / search APIs | REST + openkms-cli + skill |
| Neo4j / ontology graph | Not core | Object types, links, Object Explorer |
Teamwork Graph vs openKMS knowledge map¶
| Teamwork Graph | openKMS | |
|---|---|---|
| Purpose | Relate people, issues, pages, and connector objects for AI context | Map domain terms to channels, wiki spaces, article channels |
| Scope | Whole Atlassian + connected SaaS | Curated taxonomy + optional ontology + wiki link graph |
| Best for | “Who owns this initiative?” across Jira and Confluence | “Where is the canonical material for this term?” for KM and RAG |
openKMS does not replicate a 100B-object enterprise graph; it offers explicit, governable navigation and structured data for agents (knowledge map, ontology).
When to choose which¶
| Priority | Lean toward |
|---|---|
| Already on Atlassian Cloud; Jira-centric delivery | Confluence + Rovo |
| Must keep knowledge in your VPC / air-gapped | openKMS |
| Policy PDFs, customs, legal corpora with parse + lifecycle | openKMS |
| Cross-app search (Slack + Drive + Jira) out of the box | Confluence + Rovo |
| Prove KB answers with chunk-level debug and eval regressions | openKMS |
| Whiteboards, live docs, Remix to slides/prototypes | Confluence + Rovo |
| Articles + wiki + KB in one ACL model you operate | openKMS |
Coexistence patterns¶
| Pattern | Description |
|---|---|
| Confluence as author, openKMS as RAG plane | Sync or export pages into openKMS document/wiki channels for parsing, lifecycle, and KB Q&A (connector backlog). |
| Confluence as system of record, Rovo for daily search | Teams keep writing in Confluence; openKMS holds regulated subsets with stricter ACL and eval gates. |
| External agents | Claude/Cursor use Atlassian MCP for Confluence; openkms-skill for openKMS—different corpora, different permission models. |
RAGFlow (see comparison) already lists Confluence sync; openKMS connectors are not at parity today.
Gap analysis for openKMS¶
| Confluence AI strength | openKMS status |
|---|---|
| Unified enterprise search across 50+ apps | Global search + KB; connector sync backlog |
| Teamwork Graph context | Knowledge map + ontology; no automatic work graph |
| Rovo Studio / no-code agents | LangGraph services only; in-product agent backlog |
| Remix / multimodal republishing | Not planned as equivalent |
| Credit-based AI metering | You bring your own models; ops cost, no per-chat credits |
| Native Jira issue creation from pages | Article/document relationships; not Jira |
| openKMS strength | Confluence AI gap |
|---|---|
| Document parse pipeline + versions | Attachment-oriented; less VLM-native workflow |
| Resource ACL + admin-not-superuser | Space permissions; different compliance story |
| Evaluation module for KB/wiki | No first-class eval harness in Confluence |
| Self-hosted / Apache 2.0 stack | SaaS commitment |
Mapping to Operational Knowledge Fitness (OKF)¶
Using the OKF facets:
| Facet | Confluence + Rovo | openKMS |
|---|---|---|
| Discoverability | Strong Rovo Search + graph | Strong channel/map/search; weaker SaaS connector breadth |
| Verifiability | Citations in Rovo flows; black-box to operators | Chunk/source/debug fields; article print + versions |
| Navigability | Spaces, labels, definitions | Wiki, glossary, knowledge map |
| Currency | Page history; less policy-effective dating | Lifecycle fields, superseded RAG exclusion, impact workflow backlog |
| Reuse | High if wiki is habit; risk of private Rovo chats | Contribution paths + eval; ask→contribute backlog |
References¶
| Source | URL |
|---|---|
| Rovo in Confluence | https://www.atlassian.com/software/confluence/ai |
| Rovo product | https://www.atlassian.com/software/rovo |
| Confluence — write/edit with Rovo | https://support.atlassian.com/confluence-cloud/docs/write-or-edit-content-using-atlassian-intelligence/ |
| Remix and partner agents (blog) | https://www.atlassian.com/blog/confluence/rovo-remix-3p-agents-confluence |
| Rovo MCP gallery (blog) | https://www.atlassian.com/blog/announcements/rovo-mcp-gallery |
| Atlassian Rovo MCP Server | https://support.atlassian.com/atlassian-rovo-mcp-server/docs/use-atlassian-rovo-mcp-server/ |
| Confluence pricing (credits tiers) | https://www.atlassian.com/software/confluence/pricing |
| openKMS architecture | architecture.md |
| openKMS wiki / agents | wiki-spaces.md |
Research note written 2026-06. Atlassian feature names, credit rules, and rollout dates change frequently—validate against Atlassian Support before procurement or compliance sign-off.