Foundry MVP / V2 / V3 Roadmap

V1 lane: CTO / local-inference buyers

V1 stays in its lane. The primary buyer is the CTO or engineering lead who wants private/local AI infrastructure that works. The full stack ships together: Foundry (inference) + llm_stats (observability) + choice of OpenClaw (orchestration) or Hermes (messaging). The harness is the steering wheel, but the CTO picks the one that fits their use case — they don't need both bolted together by default.

Service-business workflow automation is a V2 path, not a V1 distraction.

Product framing

Foundry is a managed Apple Silicon private/local AI operations product for teams that want to reduce cloud AI spend, keep inference on-prem, or use private AI to reduce repetitive admin workload.

The core product stays horizontal: reliable local inference, capacity guardrails, health visibility, benchmark evidence, support runbooks, and deployment discipline.

Buyer paths:

• Primary: startup CTOs, technical founders, heads of engineering, and mid-market engineering leads who need reliable local inference.
• Secondary: service-business and ops-led buyers in trades, field services, logistics, and document-heavy operations who need private/local workflow assistance for admin-heavy processes.

OpenClaw/Hermes positioning:

• optional extension/add-on for pure CTO/local-inference buyers
• central operating/workflow layer for service-business buyers
• bounded pilot only in MVP
• reusable vertical workflow packs in V2
• stronger orchestration/fleet policy in V3

This roadmap deliberately separates:

• MVP = sellable design-partner offer
• V2 = repeatable productised deployment plus reusable vertical workflow packs
• V3 = broader platform / fleet / orchestration capability

---

MVP — Productised service

Goal

Prove buyers will pay for reliable local-inference setup, observability, and guidance.

Customer

• startup CTO / founder with existing AI spend
• one or two mid-market buyers with strong compliance or sovereignty pressure
• optionally one service-business or document-heavy operator with a narrow, repetitive admin workflow suitable for human-reviewed local AI assistance

Offer

1. Advisory pack

2. Managed setup on customer Mac Studio

3. On-prem pilot for one internal workflow

4. Optional bounded workflow pilot for a service-business buyer, using OpenClaw/Hermes only where needed

Must-have product components

• supported runtime stack definition (blessed path)
• hardware-fit questionnaire / intake
• capacity guard and safe-profile recommendations
• benchmark evidence for supported models
• llm_stats menu bar / dashboard for live visibility
• runbook for startup, restart, failure modes, and storage hygiene
• clear “local good fit / cloud better fit” guidance
• optional workflow-fit worksheet for service-business prospects
• explicit human-review boundary for any pilot involving customer comms, dispatch, finance, or system-of-record actions

Must-have operational components

• onboarding checklist
• standard install scripts
• customer environment worksheet
• support boundaries and exclusions
• update/rollback procedure
• lightweight incident log
• pilot approval notes for any workflow test: what AI may draft/classify/retrieve, what humans must approve, and what is out of scope

MVP deliverables

• PRD.md
• customer-facing landing page copy draft
• design-partner outreach list + qualification criteria
• installer/runbook bundle
• pricing hypothesis
• one demo environment on the Mac Studio
• one documented workflow pilot template, if the service-business route is tested

What MVP explicitly does NOT do

• multi-node orchestration
• auto-remediation that kills/restarts services without operator control
• billing per token/request
• full enterprise SSO/RBAC
• hardware-agnostic support matrix
• broad reusable workflow automation packs
• unsupervised customer messaging, dispatch actions, finance actions, or writes to systems of record

MVP success gate

• 3 paid design partners or 5 serious pilot conversations with clear budget/timeline
• 1 deployment used for real internal work weekly
• 1 testimonial or hard ROI/admin-time case study
• if a workflow pilot runs, 100% of operational outputs remain draft/recommendation until human approval

---

V2 — Productised deployment and vertical workflow packs

Goal

Turn the service into a repeatable offer with less founder hand-holding, and productise only the workflow patterns proven safe in MVP.

Product additions

• browser dashboard in addition to menu bar
• saved workload profiles by use case (support, summarisation, code assist, internal search, document analysis)
• runtime comparison and recommendation UI
• audit/event timeline
• guided model updates and rollback support
• explicit “no-cloud mode” configuration
• reusable vertical workflow packs for selected service-business use cases:
• job intake/triage
• quote drafting
• dispatch prep
• invoice/completion-pack admin
• customer comms drafting
• document classification/retrieval
• review queues and approval states for workflow outputs
• templates for service-business onboarding, document sources, and workflow boundaries

Technical additions

• config validation and health-check bundle
• better LM Studio / Ollama / omlx adapters
• benchmark scheduling and history views
• policy engine: what can load, what cannot, and why
• backup/export/import for state and config
• OpenClaw/Hermes packaged as an installable workflow layer where the customer buys workflow capability
• audit trail for AI drafts, classifications, retrievals, and human approvals

Logistics additions

• hardware recommendation matrix by workload and budget
• procurement guidance for Mac Studio + SSD + networking
• remote support standard operating procedure
• standard support plan and response windows
• deployment pack for customer-owned hardware
• vertical workflow discovery checklist to prevent bespoke automation sprawl

Commercial additions

• clearer packaging:
• Advisory
• Managed Setup
• Foundry Ops Retainer
• Foundry Workflow Pack add-ons
• ROI calculator: compare monthly API spend vs local TCO
• admin-time calculator: estimate repetitive turnaround reduction and avoided incremental hiring pressure without leading with headcount cuts
• first vertical-specific landing pages (legal, healthcare, finance, internal R&D, trades/services/logistics)

V2 success gate

• at least 10 paying customers or 3 retained ops customers
• onboarding time under 4 hours for supported setups
• support incidents trend downward as deployments standardise
• at least one workflow pack used weekly with documented human-review compliance and measurable admin-time improvement

---

V3 — Platform / fleet / orchestration layer

Goal

Support multiple environments, stronger orchestration, and higher-trust organisational adoption.

Product additions

• multi-machine/fleet visibility
• team dashboard with roles
• central policy and deployment templates
• alerting integrations (Slack/email/webhook)
• workflow routing between local and cloud based on policy, cost, or sensitivity
• stronger OpenClaw/Hermes orchestration across teams, queues, workflow packs, and systems
• cross-workflow memory and context controls with permission boundaries
• richer admin operations cockpit for service-business buyers

Technical additions

• managed fleet inventory
• richer audit logging and reporting
• workload tagging and usage analytics
• optional request gateway / proxy layer
• controlled failover rules
• policy-driven routing for sensitive documents and local-only workflows
• integration framework for CRM, job-management, ticketing, finance, document storage, email, and messaging systems
• guarded automation primitives with approval policies, rollback, and incident review

Logistics additions

• certified deployment patterns
• partner/reseller or managed-install channel
• formal documentation pack for IT/compliance review
• optional air-gapped deployment package
• stronger implementation playbooks for vertical workflow rollouts

Commercial additions

• enterprise plans
• annual support contracts
• compliance-specific packages
• workflow pack marketplace or partner services catalogue
• services marketplace for supported rollout partners

V3 success gate

• repeatable mid-market deployments
• at least one compliance-driven reference customer
• at least one service-business reference customer using multiple workflow packs under approval controls
• documented procurement-to-production motion that does not rely on Ged being in the loop for every install

---

Technical sequencing

Phase 1: stabilise the truth layer

• benchmark store is trustworthy
• capacity guard profiles reflect real memory behavior
• runtime probes accurately describe health and loaded models
• storage accounting catches duplicate/unsafe model sprawl

Phase 2: improve operator visibility

• menu bar stays fast and legible
• summary view explains risk clearly: healthy / caution / drift / broken
• dashboard exposes the same truth without terminal dependency

Phase 3: make deployment repeatable

• installer scripts
• config templates
• support runbooks
• rollback and update discipline

Phase 4: prove one bounded workflow safely

• choose one service-business admin workflow
• use OpenClaw/Hermes as the workflow layer only for that pilot
• draft/classify/retrieve locally
• require human review before customer, dispatch, finance, or system writes
• measure admin turnaround and review quality

Phase 5: productise workflow packs

• reusable vertical packs move into V2 after the bounded pilot proves the pattern
• review queues, approval state, and audit logs become standard
• onboarding templates prevent bespoke workflow sprawl

Phase 6: enterprise hardening and orchestration

• audit logs
• restricted/no-cloud mode
• backup/restore
• fleet view
• stronger workflow routing and policy enforcement

---

Logistics sequencing

Immediate

• define one blessed deployment target
• define one intake form
• define one pilot support boundary
• define one optional workflow pilot boundary for service-business buyers

Next

• define procurement bundle (Mac Studio / RAM / SSD / accessories / networking)
• define support offer and SLA language
• define who owns customer onboarding, support, and follow-up
• define the first V2 workflow-pack candidates from MVP evidence

Later

• standardise shipping / on-prem install playbook if hardware is supplied by us or pre-configured
• build compliance pack for regulated buyers
• build vertical workflow rollout packs for service businesses

---

Recommendation

Do not overbuild V3 platform dreams first.

The right order is:

1. Sell the core Foundry service to CTO/local-inference buyers

2. Standardise the deployment around reliable private/local AI ops

3. Prove one bounded human-reviewed workflow for the service-business buyer path

4. Productise repeatable vertical workflow packs in V2

5. Only then build stronger orchestration, fleet, and policy capabilities in V3