Roadmap

At a glance

2026 H2               2027 Q1                  2027 Q4                 2028+
  │                     │                        │                       │
  v1.0 GA               v2.0 MVP                 Track B MVP             Verification Engine
  Pilot cohort          Knowledge Fabric         Jira · Confluence       PR gates · blast radius
  Git + code (today)    Memory · CHRONOS · OIR   SharePoint · Git · docs SSO at scale

Capabilities and order evolve with design partner feedback. We label shipping vs planned explicitly — no certification or autonomous-governance claims.

v1.0 · end of August 2026

Decision-control plane

Local-first decision-control plane for software teams — evidence before inference, human approval before repo changes. In internal pilot now; design partner intake opens by start of September 2026. General availability end of August 2026.

ContextMint Engine — tray-native operator console (server, Ollama, models, indexing, quality scans, logs, support bundle)
VS Code extension — Ask / Plan / Agent modes, context lanes (Repo / Work / Hybrid), chat sessions, attach context, suggested prompts
Context Lens (gate 1) — preview retrieved chunks, scores, routing, and governance notes before the model runs
Patch preview (gate 2) — review unified diffs before anything applies; no silent writes
Quality map — stored governance findings (clones, near-duplicates, structure signals) with optional apply gates
Federated query router v0.1 — governance lookup reads the Findings Store first (store-first, tier-agnostic facts, trust metadata) — not chat-time rediscovery
Context packs — per-repo manifests with @pack:name activation and auto-sync
Glass-box routing — provenance breakdown, routing badge, reasoning trace for power users
Conventions learner — proposals and audit history with human approval before rules enter the graph
Image evidence — paste UI screenshots in chat; local Ollama vision by default; Context Lens preview before send
ARGUS visual audit — structured WCAG-tagged findings on screenshots; report only, human-gated fixes
Workspace-agnostic evidence assembly — index-time route knowledge when available, then hybrid search + expand (file windows, grep, dependency graph) on any indexed repo
Symbol-aware retrieval — citations name file, symbol, and line range — not anonymous text chunks
Local-first default — Engine + Ollama on loopback; code stays on device

Optional (off by default): BYOK cloud routing, git sandbox worktrees, inline completion, shared on-prem API server for enterprise pilots — full v1.0 list.

Recruiting design partners

2026 pilot cohort

A 30-day evaluation pilot is a bounded proof-of-concept — not a product subscription limit. After onboarding, 10–30 developers on one team or repo use ContextMint in real work for one calendar month; we review together and deliver a written summary for security and platform leadership. Extendable to 90 days when the pilot is going well.

You get direct engineering access; we get honest signal on fit for regulated and sovereign AI dev workflows — and priority input on v2.0 and Track B (including which knowledge systems to connect first).

Guided pilot playbook — week-2 and week-4 feedback (async OK)
Deployment path review — local Engine per developer or shared on-prem server (deployment models)
Security questionnaire support — architecture summaries for stakeholders
Priority input — influence v2.0 Fabric scope and Track B enterprise packaging
Team rollout patterns — Docker / Helm shared-server pilots with platform engineering
Pilot summary report — outcomes for security and engineering leadership

Design partner details Book a demo

v2.0 · target Q1 2027

Engineering Intelligence Platform

After v1.0 exit, ContextMint becomes one living digital twin of the engineering system — not seven siloed tools. Retrieval, memory, health analysis, temporal forensics, and multimodal validation become query projections over a single logical model: the Knowledge Fabric. Starts post–v1.0 freeze; target MVP first quarter of 2027 (Q1).

Six waves (post–v1.0)

Wave 1 — Institutional memory

Fabric shadow-write, learner approval gate, Memory Tab — conventions become graph nodes with evidence.

Wave 2 — Temporal & structural

CHRONOS checkpoint forensics, health analysis, Architecture Tab — graph/metadata first, not commit embeddings.

Wave 3 — Zero-latency UX

ORACLE prefetch — intent + retrieval context cached before you ask; extends v1 PERF layers.

Wave 4 — Multimodal validation

ARGUS audit reports in sandbox — structured findings, human reviews fixes separately.

Wave 5 — Distributed compute (lite)

MESH-Lite — explicit trusted peers for heavy VLM/health jobs; no auto-discovery swarm.

Wave 6 — conditional

Full mesh only if Wave 5 proves measurable runtime benefit in dogfood.

Operational intelligence (within v2.0)

v1.0 assembles named engineering artifacts with index-time knowledge objects (routes today), hybrid search + expand, and human gates. v2.0 adds scoring belief across more object kinds — and reconciles conflicting signals into one auditable answer.

Knowledge reconciliation — when Git, CODEOWNERS, team memory, and docs disagree, surface one scored claim with supporting and conflicting evidence
Operational intelligence retrieval (OIR) — rank by relevance and trust, freshness, and evidence coverage
Domain-aware retrieval — files and symbols tagged to domains; filter by ownership and architectural boundaries, not chunks alone
Repository DNA + Feature Registry Lite — deterministic health baselines and feature traceability on the Fabric
Summary-first retrieval — semantic compression at index time; expand to raw code only when confidence is low
RBAC + audit foundations — tenant_id / owner_id on Fabric nodes; immutable audit schema (full enterprise packaging continues in Track B)

We do not claim trust-weighted retrieval or cross-system reconciliation until product gates pass. v1.0 already ships index-time route knowledge, store-first governance lookup, and symbol-aware citations. Glossary →

Track B · target end 2027

Enterprise & organizational knowledge

Track B extends ContextMint from developer code intelligence into an explainable organizational knowledge platform — same retrieval core and local-first posture, with security and compliance controls that regulated buyers require. Builds on the v2.0 Fabric; target MVP end of 2027. After design partner intake opens, cohort feedback will help prioritize which external systems to connect first.

Connected knowledge (Track B): One governed index across the systems your teams already use — not code in isolation. v1.0 indexes Git repos today; Track B adds adapter-based ingest for Jira (issues, epics, ownership), Confluence (wikis, ADRs, runbooks), SharePoint (document libraries, policies), Git (multi-repo and org-wide code graph), and other knowledge bases (Notion, internal portals, ticket systems — partner-driven priority). Answers cite the source system with ACL respected per document.

Security-first build order

Foundation controls (B0) — unified authorization on every read path; cache keys scoped to tenant + role; explainable score breakdown fields
Security boundary (B3) — document-level ACL enforcement; hard multi-tenant isolation (scoped storage roots, per-tenant keys, no cross-tenant joins)
External knowledge adapters (B1+) — Jira, Confluence, SharePoint, Git org-graph, and partner-selected knowledge bases; async ingest queue with ACL from each source
Document ingestion (B1) — PDF, DOCX, and exported wiki pages with async queue, backpressure, and quarantine for OCR-heavy workloads
Metadata + PII (B2) — policy-driven redaction modes; metadata boosts bounded so base retrieval ranking stays stable and explainable
Enterprise surface (B4) — OIDC/SAML SSO, API gateway policies, admin console for roles, retention, index health, and audit browsing
Compliance packaging (B5) — GDPR deletion proof, air-gapped Helm/Docker bundles, tamper-evident audit (hash-chain + signed checkpoints), SOC 2 evidence collectors

Outcomes for design partners

Ask questions across code + Jira + Confluence + SharePoint — one answer with cited evidence, not four conflicting snippets
One platform for engineering and internal docs — same Context Lens and human-gate model
Team server deployment with SSO — extension connects to shared on-prem API, not loopback-only
Compliance exports your security team can run without vendor workshops
No cloud egress unless explicitly enabled per workspace or tenant

We do not claim live Jira/Confluence/SharePoint connectors, SOC 2 certification, or FedRAMP until gated. Adapter order follows design-partner demand; v1.0 pilots stay Git-first. Security use cases → · Co-shape Track B →

Beyond v2.0

Verification Engine

Longer-horizon capabilities on top of the Fabric — deterministic graph evidence first; AI narrates with citations, never overrides graph truth without human confirmation. Timing evolves with design partner feedback; many items follow Track B (end 2027) and extend into 2028+.

PR verification gates — ownership, impact, and policy checked on every change (vision)
Blast radius — graph traversal for impacted services, files, and teams (learn more)
Repository DNA — deterministic repo health and convention baselines (learn more)
Decision graph — architecture intent and ADR traceability (learn more)
Drift detection — scorecards for accepted vs violation baselines (learn more)
Agent governance — policy-scoped tool contracts for AI agents in CI (learn more)
ARGUS insights tab — cross-session audit history and team dashboards (beyond v1 paste + sandbox)
Enterprise CLI packaging — graph queries and verification in CI pipelines
MCP server for agents — blast radius, features, and decisions as tool contracts (contracts)
CI/CD deployment graph — pipeline and release context linked to code changes (after Track B knowledge fabric)

Explore the Verification Engine Terms explained

Capabilities and order may change as we learn from pilots. What ships in v1.0 is documented on Features. Governance and compliance claims follow our trust center — we do not promise autonomous governance or certification.