Roadmap: Open-Core Packaging and Pricing Strategy

This brief is [OWNER]-gated end-to-end. Pricing, license splits, tier definitions, target segment, and any commercial commitment are owner decisions. Agents working this brief prepare decision inputs — verified market data, unit-economics models, license-mechanics comparisons — and stop at a decision memo. No agent produces a binding design (a price, a license text, a published tier) from this page.

Gap statement: no packaging at all, in a market with clear price anchors

Wikantik is an Apache 2.0 codebase with no commercial packaging: no tiers, no hosted offering, no support contract, no published pricing. That is fine for an open-source project; it is disqualifying for a service. This brief is deliberately last-sequenced — the monetizable features are exactly the Tier 1–2 roadmap items, so packaging follows capability, not the reverse. It exists now so that capability designs keep the packaging seam in mind.

Market grounding: the anchors

Per-seat products: Onyx cloud ~$20/user/month; Glean ~$50/user/month with $60K+ annual minimums (Onyx enterprise RAG platforms).
Transparent ladders as a weapon: Ragie publishes a $0/$100/$500/Enterprise ladder and actively poaches Vectara customers with a free month; Pinecone's costs escalate sharply past 100M vectors, driving self-host interest (Atlan enterprise RAG comparison).
Usage-based: hyperscalers price per-query/per-capacity (AWS offers predictable per-query pricing for its agentic retriever; Azure AI Search bills tiered hourly capacity even at zero queries) (Bedrock KB, Azure guide).
The proven open-core seam: Onyx keeps connectors open but gates document-level permission sync to the paid Enterprise Edition (Onyx docs) — the market has already validated which layer enterprises pay for.
Market size/context: enterprise RAG ~$1.94B (2025) projected to ~$9.86B by 2030; widely-cited figure that ~95% of GenAI pilots fail to reach P&L impact — both from vendor-adjacent sources, treat as directional (Onyx insights).

Current Wikantik state

Apache 2.0 across the board — explicitly marketed as a differentiator vs GPL/AGPL wikis; any packaging must not poison that.
Cost structure of one tenant is unmeasured (see RoadmapMultiTenancyPosture — the per-tenant cost floor investigation feeds directly into pricing).
No commercial entity, no billing infrastructure, no support SLA machinery. Single-maintainer development pace is the honest constraint on any support commitment.

Proposed direction

Open-core split, following the validated seam: core stays Apache 2.0 (wiki, retrieval, bundle, MCP servers, ontology, base connectors). Paid enterprise modules: connector ACL sync (RoadmapExternalAclInheritance), fleet management / control plane (RoadmapMultiTenancyPosture), retrieval-quality dashboard (RoadmapRetrievalEvaluationObservability), retrieval audit/compliance pack (RoadmapCompliancePosture).
Hosted single-tenant tier ("managed silo"): a provisioned dedicated instance per customer — the multi-tenancy posture is the hosted product. Priced flat per instance + size band, not per seat, undercutting per-seat incumbents for small teams.
Publish a transparent ladder early (Ragie's lesson): even 'free self-host / $X managed / talk-to-us enterprise' beats silence, and forces the cost-floor homework.
Support/services tier for self-hosters as the lowest-machinery first revenue.

Agent-executable slice and acceptance criteria

There is no autonomous "Phase 1 implementation" for this brief. The agent-executable slice is decision-input preparation:

A verified pricing snapshot: re-check every anchor above against current public pricing pages (they churn fast) and record source + date per figure.
A unit-economics model: the per-tenant cost floor from RoadmapMultiTenancyPosture's [MEASURE] work + provisioning labor → minimum viable managed price, compared against per-seat anchors for a 10-seat team.
A license-mechanics comparison: separate commercial modules (Onyx model) vs license-key gating in one artifact, with fork-friendliness and Apache-2.0-purity implications laid out.

Done when: the three inputs exist as a decision memo presenting options with trade-offs — and the memo explicitly does not recommend a single price. The owner decides from it.

Verify before designing on these

Every number in the market-grounding section is a verify-first claim: the per-seat anchors, the Ragie ladder, the market-size projection, and the "95% of pilots fail" figure are all vendor-adjacent and dated 2026-07. The pricing-snapshot task above is the verification.

Non-goals

No published pricing, license text, or tier commitment from agent work — owner-only.
No billing/entitlement implementation until a first paying customer shape exists.
No relicensing of anything currently Apache 2.0 (invariant #8 in RoadmapTargetArchitecture: the core stays complete and free).

Investigation starting points

[MEASURE] Per-tenant unit economics (input 2 above — depends on the tenancy cost-floor measurement).
[AGENT] License-mechanics research (input 3 above) — mechanics only; [OWNER] chooses.
[OWNER] Which segment first: regulated self-hosters who cannot use SaaS (plays to silo + compliance strengths) vs agent-tooling developers (plays to MCP-first + SDK strengths)?
[OWNER] Billing/entitlement minimalism, commercial-entity formation, trademark and brand separation between the OSS project and the commercial offering (wikantik.com marketing site exists).