Where the graph layer
goes next.

One schema per department.

Each schema defines how a kind of spec is structured into typed assertions. Engineering Spec is the default; Policy Shield and Incident Record cover the security and operational facts agents get wrong most expensively.

The default schema is documented at /schemas.

Graph you can govern, not just query.

The milestone that turns MsgStack into a governance layer: score any content against the graph, gate generation on approval status, and give every claim a named owner. Much of it has already landed ahead of the milestone.

• Alignment scoring — per-section 0–100 reports with hard vs soft conflict classificationShipped
• Approval-gated grounding — generation skips non-approved entries by defaultShipped
• Content tiering — per-entry contract from Tier 1 verbatim to Tier 3 groundedShipped
• Cross-spec traversal — k-hop expansion that leaves the spec it started inShipped
• Typed graph edges — DEPENDS_ON, INFORMS, SUPERSEDES, CONTRADICTS, OWNS, IMPLEMENTSShipped
• Hybrid fusion — reciprocal rank fusion of vector recall with graph expansionShipped
• ·Dual output — citation-marked review copy alongside the clean deliverablePlanned
• ·Self-service portal — shareable generation link, no admin account requiredPlanned

Beyond marketing, into every domain of truth.

Typed edges across specs, cross-spec traversal, and change propagation — the graph doing work that vector search cannot.

• ·engineering_spec schema — API constraints, SLAs, deprecation noticesPlanned
• ·policy_shield schema — legal language retrieved verbatim, never paraphrasedPlanned
• Sub-canons — nested domains with four configurable inheritance policiesShipped
• ·Cross-domain dependency edges — INFORMS and DEPENDS_ON relationshipsPlanned
• Change propagation — transitive staleness along DEPENDS_ON and INFORMSShipped
• ·Content CI/CD pipeline — validate, test, merge, propagate, auditPlanned
• ·Automatic entity extraction — propose MENTIONS edges from assertion textPlanned
• ·Identity-scoped retrieval — SSO login and per-user embargo scopingPlanned

More sources, sharper grounding.

Broaden what the graph can ingest — decks, spreadsheets, audio — and bring competitive intelligence into the graph so runbooks counter what competitors actually say.

• PPTX and XLSX extractionShipped
• Ingestion conflict detection — uploads flagged against existing graph before commitShipped
• ·Render-mode tagging — verbatim-insert vs parse-as-content per assetPlanned
• ·Industry and segment variants — audience × channel × industry retrievalPlanned
• ·Deck indexing and presentation assembly from approved slidesPlanned
• ·Audio and video indexing with timestamped moment retrievalPlanned
• ·Competitor document import and competitive gap analysisPlanned

From generated to published.

Close the gap between an approved artifact and a live asset: push grounded content into the systems where GTM teams already work, and pull published assets back in for alignment scoring.

• ·HubSpot — email drafts, social queue, and asset pull-back for scoringPlanned
• ·LinkedIn — publish social cards to company pages and profilesPlanned
• ·Salesforce — approved messages and runbooks as opportunity snippetsPlanned
• ·Google Docs export into a designated Drive folderPlanned
• ·Slack app — /msgstack generate, grounded content in the channelPlanned
• ·Outbound webhooks — send any artifact to any external systemPlanned

Agents over the graph, and a hosted edition.

Specialized agents for governance, voice, and structure sit over the pipeline, a conversation-first navigator answers questions about the graph — and a managed hosted edition opens MsgStack to teams that don't want to self-host.

• Specialized agents — governance, brand voice, and narrative structure analysisShipped
• Graph Navigator — conversation-first dashboard agent with streaming chatShipped
• Tonal sliders and controlled vocabulary filters on generationShipped
• ·Hosted SaaS mode — managed database, index, and serverPlanned
• ·Onboarding wizard and industry starter templatesPlanned
• ·Recommendation routing — agents suggest where new material fits the graphPlanned
• ·Intent-based routing — the right model tier and skill chosen per queryPlanned
• ·Content analytics — usage heatmaps, acceptance signals, value reportingPlanned

The wider constellation, loosely charted.

Later milestones and the unscheduled backlog — source connectors, advanced graph operations, and platform ecosystem work. Sequencing here depends on community feedback more than anywhere else on the map.

• Google Drive sync — OAuth connector and background sync loopShipped
• ·OneDrive, SharePoint, Notion, and Confluence connectorsPlanned
• ·Audience coverage analysis and cross-domain graph queriesPlanned
• ·Neo4j migration path for high-scale graphsPlanned
• ·Cross-domain search — one question across every department's graphPlanned
• ·Multi-LLM support — Claude, Gemini, and local models via OllamaPlanned
• ·CLI tool and VS Code extension for inline claim suggestionsPlanned