Foundgine

Roadmap

Home → Reference → Roadmap

The roadmap is intentionally vertical-slice driven. Foundgine should prove the core thesis before expanding into multiple transports, databases, retrieval systems, or AI integrations.

Phase 0 — Execution substrate

Status: proven in the canonical sample

Domain
→ Metadata
→ Dynamic Planner
→ QueryPlan
→ ProviderPlan
→ SQL
→ SQLite
→ Result

The Banking sample is the acceptance test.

Phase 1 — Semantic domain

Status: next

Create a protocol-neutral semantic model representing:

• entities
• identities
• fields
• relationships
• searchable properties
• actions
• policies

Start hand-authored if necessary.

The goal is to prove the model before building a compiler.

Phase 2 — Resolution

Status: planned immediately after Phase 1

Resolve human/agent references into domain identities.

Required behavior:

• exact matches
• useful search fields
• relationship traversal
• ambiguity reporting
• evidence for why an entity was selected

No silent guessing.

Phase 3 — Read planning

Status: planned

Convert an intent into a Foundgine query plan.

Intent
→ Resolve
→ Semantic query
→ QueryPlan
→ ProviderPlan
→ Execute
→ Evidence

The first test is the Banking “last five transactions” scenario.

Phase 4 — Domain actions

Status: planned

Introduce explicit, constrained action descriptors.

IssueRefund
SuspendAccount
ChangeTier

Agents may select declared actions; they cannot invoke arbitrary CLR methods.

Phase 5 — Policy and authorization

Status: planned

Make authorization a planning input rather than a late controller concern.

Intent
→ Resolve
→ Policy
→ Plan

The result should explain allow/deny decisions.

Phase 6 — Preview and approval

Status: planned

Mutations become:

Plan
→ Preview
→ Approve
→ Execute

Preview is part of the execution contract.

Phase 7 — Verification and evidence

Status: planned

Every important mutation should verify expected state after execution and produce an evidence chain.

Phase 8 — MCP

Status: planned

MCP becomes a thin adapter over the semantic API.

Initial surface:

discover
resolve
plan/query
preview
execute
evidence

Do not create an entity-specific tool for every domain operation unless a concrete integration proves it necessary.

Phase 9 — More execution targets

Status: later

Add execution targets behind the existing plan:

Structured data
Domain actions
Semantic retrieval
External data

Do not build all targets in parallel.

Phase 10 — Compile-time semantic compiler

Status: later

Use Roslyn to derive and generate:

• stable IDs
• entity metadata
• relationship metadata
• search descriptors
• action descriptors
• policy metadata
• planner hints

The compiler describes the legal application vocabulary.

It does not attempt to generate future natural-language plans.

Phase 11 — Ecosystem integrations

Potential integrations:

• ASP.NET Core
• MCP
• Semantic Kernel
• OpenTelemetry
• EF Core
• Dapper
• Temporal
• Kafka
• PostgreSQL/pgvector
• other databases and retrieval systems

These remain adapters/integrations rather than the Foundgine core.

Explicit non-goals

The roadmap does not include building:

• a proprietary LLM
• a general agent framework
• a proprietary vector database
• a replacement for MCP
• a replacement for EF Core
• a replacement for Temporal
• a replacement for Kafka

The release gate

Do not call the first AI-native milestone complete until a real test demonstrates:

natural-language request
→ domain resolution
→ policy
→ plan
→ real execution
→ verification
→ evidence

for both a read and a mutation.

This site is open source. Improve this page.