Product Roadmap — Contextful · Meerkat · Spore
What the products are is in flagship products; how they compose from packages is in product stack. This is the build plan — what ships when across H2 2026, and why in that order. It’s the product half of the h2 2026 operating plan and the concrete expression of the bet Observation 3 (unmet need is agent guardrails + FDEs → Meerkat is the spearhead) and Observation 4 (local-first → Contextful is single-binary, data never leaves the boundary).
Sequencing rationale
- Contextful now. The launching product and the shared engine (Hakiri data plane) the others build on. Local-first, single-binary, MCP-native — the wedge for the underserved local-first market.
- Meerkat in a parallel Q3 block. Package boundaries are disjoint from Contextful (
@fractalboxdev/meerkat-*governance plane vs@fractalboxdev/hakiri-*data plane; only the MCP surface, capability tokens, and audit trail are shared), so it builds alongside Contextful in the same 13-week sprint without contention. It’s the governance/control-plane moat and the front-door wedge (productized assessment → continuous governance). - Spore after Contextful traction. Same engine, different surface — multi-tenant, embedded in a vendor’s product. Demand-gated: build the embed SDK only once Contextful has proven the core with known teams.
The Q3 dev block (locked)
One 13-week sprint, 2026-07-01 → 2026-09-30 builds Contextful and Meerkat in parallel to a demo-ready gate, feeding a Q4 lead-wedge launch.
Milestones
| When | Milestone | Product |
|---|---|---|
| 2026-07-01 | Q3 dev block starts (13w, locked) | Contextful + Meerkat |
| 2026-08-15 | Mid-sprint review / demo | Both |
| 2026-09-30 | Demo-ready gate (GA-ready) | Both |
| 2026-10-01 | Lead-wedge launch (assessment → governance) | Meerkat-led |
| 2026-11-15 | Contextful GA | Contextful |
| 2026-11-30 | Meerkat GA | Meerkat |
| Q4 → traction-gated | Spore embed SDK build begins | Spore |
What each ships as
- Contextful — single-binary local-first runtime; WASM-sandboxed connectors, MCP-native context store (Parquet + SQLite catalog, DuckDB-queryable), capability tokens, object-storage sync. Revenue: sell the sync, not seats. First design partners: AX, Numu, Vascue (team).
- Meerkat — deterministic use-case guardrails, permissions/scoping
(agent, host, on-behalf-of, task-id), replayable regulator-legible audit trail, compliance mapping (AI Verify, EU AI Act, SOC 2). Open-source the probe client + cage via OpenHackersClub (community). - Spore — multi-tenant, isolated, embeddable agent context for a vendor’s own product (“Workato Embedded for agent context”). Per-tenant pricing. First fit: Vascue (embed).
Dogfood link
Every one is dogfooded before it reaches a client — the harness, Hakiri engine, and Meerkat cage run our own org first. That discipline is dogfood. The runway caveat — the Q3 block needs a bridge, since grant revenue doesn’t land before ~Q1 2027 — is tracked in roadmap and the h2 2026 operating plan.