Sokrates Roadmap

From current implementation to full product vision. Organized by dependency chain, not calendar.

See also: Sokrates Implementation Status for what’s shipped today. Technical Architecture Whitepaper for the full vision. Product Description and Feature Set for the product-level spec.

How to Read This

Epics are grouped into phases. Phases have dependency relationships — Phase 1 work generally depends on Phase 0 foundations. Within a phase, epics are independent unless noted. Each epic links to the wiki page(s) that define the target architecture and to the code location where implementation will land.

Phase 0: Foundational Infrastructure

Prerequisites that unlock everything downstream. These are the load-bearing gaps between “what’s built” and “what the product needs.”

E0.1: DMCG Schema Ingestion Pipeline

The single highest-leverage unbuilt system. Everything in the product vision — automated data hygiene, schema healing, zero-configuration integration — depends on this.


What	OpenAPI spec in, typed Hyle `BaseNode` subclasses out, registered in `NodeRegistry`, queryable immediately
Wiki refs	Technical Architecture Whitepaper SS2, Hyle Graph-Native ORM with Dynamic Schema Registry, Hyle Design Lineage (Grimoire to Hyle)
Code target	`eidos/src/eidos/infrastructure/schema_ingestion/` (new)
Depends on	Nothing — `BaseNode`, `HyleMeta`, `NodeRegistry`, `@node` decorator are all shipped
Unlocks	E0.2 (schema healing), E1.1 (customer onboarding), E2.1 (semantic enrichment), E3.2 (fleet basis learning)

Scope:

Add datamodel-code-generator dependency to eidos/pyproject.toml
Build ingestion service: accepts OpenAPI spec (string, URL, or file path) → calls DMCG Python API with --base-class hyle.BaseNode flags → writes module to disk with version-namespaced path
Build loader: importlib loads generated module → HyleMeta fires during class creation → types auto-register in NodeRegistry
Build API endpoint: POST /ingest accepts OpenAPI spec, returns registered type names and count
Integration test: round-trip from OpenAPI spec → registered types → Neo4j persistence → query via unified DSL

E0.2: Schema Healing (Hot-Swap)

Makes DMCG ingestion continuous rather than one-shot.


What	Detect schema changes, regenerate models, hot-swap via `importlib.reload()`, update Neo4j constraints
Wiki refs	Technical Architecture Whitepaper SS2.3, Hyle Graph-Native ORM with Dynamic Schema Registry
Code target	`eidos/src/eidos/infrastructure/schema_ingestion/healing.py` (new)
Depends on	E0.1 (DMCG pipeline)
Unlocks	Self-healing data integration — the core product promise

Scope:

Schema change detection: poll or webhook for OpenAPI spec changes, diff against stored spec hash
Version-namespaced regeneration: hyle.generated.customer_v{N+1} module naming
Hot-swap: remove old module from sys.modules, load new via importlib, HyleMeta re-registers
Old version preserved on disk for audit and rollback

E0.3: Eidos Observer Wiring

Connect the NodeRegistry.on_register() hooks to actual Eidos infrastructure.


What	When a new node type registers, Eidos auto-creates Neo4j constraints/indexes, updates schema cache, optionally triggers relationship inference
Wiki refs	Technical Architecture Whitepaper SS1.2
Code target	`eidos/src/eidos/main.py` (lifespan), `eidos/src/eidos/infrastructure/neo4j/`
Depends on	Nothing (hook mechanism is shipped in `hyle.py:74`)
Unlocks	E0.1 becomes truly zero-configuration (types register and Neo4j is ready automatically)

Scope:

In Eidos lifespan startup, subscribe to _default_registry.on_register()
Callback creates Neo4j unique constraint on id for the new label
Callback creates any _graph_indexes declared by the @node decorator
Callback updates an in-memory schema cache for the unified query DSL
If the type has an embedding field, ensure it’s included in the vector index

E0.4: Curator End-to-End Pipeline

Wire the existing curator agent into a fully operational loop.


What	Cron-triggered curator runs → CurationReport → Telegram/Discord approval → action execution
Wiki refs	Eidos Curator Agent and Hyle Ontological Framework
Code target	`agents/src/agents/curator/`, `hermes-plugin/sokrates/`, Hermes cron config
Depends on	Nothing (curator agent, models, EidosClient, and Hermes plugin all exist)
Unlocks	Autonomous graph maintenance — the Sókrates Agent can trust Eidos data quality

Scope:

Configure Hermes cron job to invoke run_curator daily
Surface CurationReport.summary via Hermes channel (Telegram/Discord)
Implement approval flow: human replies approve/reject proposals
Build action executor: executes approved merges, archives, deletions via Eidos API
Log all actions as Observation nodes in the graph (audit trail)

Phase 1: Product Launch Readiness

Features needed for first customer engagement. Assumes Phase 0 foundations are in place.

E1.1: Customer Onboarding Pipeline

The end-to-end flow for connecting a new customer’s systems.


What	Automated flow: customer provides API credentials → DMCG ingests their schemas → Eidos indexes → Sokrates Agent has a queryable graph of their operations
Wiki refs	Product Description and Feature Set SS4.1, Go-to-Market Strategy SS6.1
Code target	`eidos/src/eidos/services/onboarding/` (new)
Depends on	E0.1 (DMCG), E0.3 (observer wiring)
Unlocks	First customer deployment, EDIH-IS trial, the “14-day audit” GTM motion

E1.2: Inefficiency Invoice (Monthly Value Report)

Automated monthly report quantifying what the Sokrates Agent found and fixed.


What	Monthly report: structural friction discovered, estimated cost, remediation status. PDF or structured markdown, delivered via Hermes channel.
Wiki refs	Gemini Deep Think analysis (scratchpad.md), Revenue Model and Pricing (churn prevention)
Code target	`agents/src/agents/reporter/` (new)
Depends on	E0.4 (curator producing observations), metalayer or at minimum basic topology queries
Unlocks	Churn prevention, value demonstration to CEO buyer, expansion revenue conversations

Scope:

Monthly aggregation of curator observations, anomalies, and remediation actions
Cost estimation heuristics (configurable per customer: average hourly rates, workflow frequencies)
Template-driven report generation (structured markdown → PDF via weasyprint or similar)
Delivery via Hermes channel to customer’s designated contact
Historical report storage as Observation nodes in customer’s Eidos graph

E1.3: Auto-Generated Claude Skills from Operational Topology

The Sokrates Agent discovers workflows and writes Claude Skills that encode them.


What	Agent observes operational topology via Eidos graph → generates `SKILL.md` files → deploys to customer’s Claude workspace
Wiki refs	Product Description and Feature Set (plugin design mode), Gemini Deep Think analysis
Code target	`agents/src/agents/skill_generator/` (new)
Depends on	E0.1 (DMCG — needs typed nodes to reason about), E0.4 (curator — needs clean graph)
Unlocks	The core product loop: Sokrates Agent discovers friction → generates automation → customer uses it in Claude

E1.4: Hardware Tiering (Standard / Sovereign)

Two deployment targets based on sovereignty needs.


What	Standard (GMKtec/Beelink, API-routed inference) and Sovereign (DGX Spark, fully local inference) NixOS configurations
Wiki refs	Technical Architecture Whitepaper SS5, Gemini Deep Think analysis
Code target	`nixos/` (extend existing configs)
Depends on	E2.2 for Sovereign tier (Gemma must run locally); Standard tier depends on nothing new
Unlocks	Broader addressable market — sovereignty-sensitive and cost-sensitive customers both served

Note: Standard tier is essentially what’s already shipped (GMKtec dev box + API inference). Sovereign tier requires the Gemma integration work in Phase 2.

Phase 2: Intelligence Layer

On-device inference, semantic enrichment, and the metalayer. The transition from “knowledge graph with connectors” to “autonomous organizational intelligence.”

E2.1: Gemma 4 Semantic Enrichment

On-device model for schema enrichment, ontological classification, and entity resolution.


What	Gemma 4 31B Dense model running locally on DGX Spark, performing semantic field enrichment, auto-classification into four primitives, and cross-source entity resolution
Wiki refs	Technical Architecture Whitepaper SS3
Code target	`eidos/src/eidos/infrastructure/enrichment/` (new)
Depends on	E0.1 (DMCG — needs generated types to enrich), DGX Spark hardware
Unlocks	E2.3 (metalayer — needs enriched types), fully automated onboarding

E2.2: Local Inference Stack (DGX Spark)

NixOS configuration for running Gemma on the DGX Spark.


What	NixOS module for DGX Spark hardware, model serving (vLLM or similar), integration with Eidos
Wiki refs	Technical Architecture Whitepaper SS5.1-5.2
Code target	`nixos/modules/`, `nixos/hardware/`
Depends on	DGX Spark hardware availability
Unlocks	E1.4 Sovereign tier, fully sovereign deployment, on-device LoRA fine-tuning

E2.3: Hypergraph Metalayer (MVP)

Generating queries that compute organizational topology.


What	Named queries whose result sets define hyperedge membership. Materialized/virtual distinction. Basic stratification.
Wiki refs	Technical Architecture Whitepaper SS4, Hyle Graph-Native ORM with Dynamic Schema Registry (hypergraph semantics)
Code target	`eidos/src/eidos/domain/metalayer/` (new)
Depends on	E0.1 (DMCG — needs ground facts), E2.1 (enrichment — needs classified nodes) for full value; can prototype against existing hand-crafted nodes
Unlocks	E1.2 (Inefficiency Invoice — topology-driven cost estimation), the “always-on organizational intelligence” product promise

Scope:

Metalayer DSL: define generating queries with MATERIALIZED/VIRTUAL annotation
Evaluation engine: execute generating queries, cache materialized results, recompute virtual on demand
Stratification: prevent circular dependencies between generating queries
Integration with unified query DSL: hyperedge results queryable through existing /query endpoint
Prototype differential evaluation (recompute only affected hyperedges when ground facts change)

E2.4: GQL Query Builder

Portable query compilation targeting both Cypher and GQL.


What	AST layer in the query builder that serializes to Cypher (Neo4j) or GQL (ISO 39075) via dialect drivers
Wiki refs	Technical Architecture Whitepaper SS4.6, Hyle Graph-Native ORM with Dynamic Schema Registry
Code target	`eidos/src/eidos/infrastructure/neo4j/query_builder/` (extend existing)
Depends on	Nothing — existing Cypher builder provides the foundation
Unlocks	Microsoft Fabric portability, future database migration flexibility

Phase 3: Scale and Fleet Intelligence

Cross-deployment learning, fleet management, and the compounding flywheel.

E3.1: Self-Evolution Harness

DSPy/GEPA integration for systematic skill and prompt optimization.


What	Wrap Hermes skills as DSPy modules, apply GEPA optimization, evaluate against real session history, deploy improved versions
Wiki refs	Technical Architecture Whitepaper SS5.3
Code target	`agents/src/agents/evolution/` (new)
Depends on	Sufficient session history from real customer engagements
Unlocks	Autonomous agent improvement, metalayer query evolution

E3.2: Cross-Fleet Basis Learning

Pattern sharing across deployments without sharing customer data.


What	Structural intelligence (schema patterns, ontological classifications, workflow templates) extracted from customer deployments and aggregated into the shared basis
Wiki refs	Technical Architecture Whitepaper SS9, Revenue Model and Pricing (basis as moat)
Code target	Fleet management infrastructure (TBD)
Depends on	E2.1 (enrichment — patterns must be abstracted from customer data), multiple live deployments
Unlocks	The flywheel: each deployment makes the next faster and more accurate

E3.3: Fleet Command (DGX Station GB300)

Central infrastructure for cross-fleet model training and basis aggregation.


What	Central DGX Station running larger model, producing improved LoRA adapters, distributing basis updates to edge boxes
Wiki refs	Technical Architecture Whitepaper SS5.6
Depends on	Fleet scale (30+ deployments), E3.2 (basis learning)
Unlocks	Trillion-parameter MoE for complex inference, centralized training

Research Track

No committed timeline. Informs future architecture but no dependency chain leads here.

Topic	Wiki Ref	Notes
Cognitive architecture / strange loop	Technical Architecture Whitepaper SS5.4	Patterns observing and selecting for better patterns
Metacognitive daemons	Technical Architecture Whitepaper SS5.4	Failure-mode detection from execution traces
Negative space compression	Technical Architecture Whitepaper SS5.3	Efficiently eliminating unproductive regions of idea-space
Riemannian Intent Manifold	Riemannian Intent Manifold From Philosophy to Empirical Research Program	Mathematical framework for intent-space geometry

Dependency Graph

Phase 0 (Foundation)
  E0.1 DMCG Pipeline ─────────┬──→ E0.2 Schema Healing
  E0.3 Observer Wiring ────┐  │
  E0.4 Curator Pipeline    │  │
                           │  │
Phase 1 (Product Launch)   │  │
  E1.1 Onboarding ←───────┴──┘
  E1.2 Inefficiency Invoice ←── E0.4
  E1.3 Auto-Generated Skills ←── E0.1, E0.4
  E1.4 Hardware Tiering (Standard = ready; Sovereign = needs E2.2)

Phase 2 (Intelligence)
  E2.1 Gemma Enrichment ←── E0.1, DGX Spark hardware
  E2.2 Local Inference Stack ←── DGX Spark hardware
  E2.3 Metalayer MVP ←── E0.1, E2.1 (or prototype without)
  E2.4 GQL Query Builder ←── nothing (extend existing)

Phase 3 (Scale)
  E3.1 Self-Evolution ←── session history from live deployments
  E3.2 Fleet Basis Learning ←── E2.1, multiple live deployments
  E3.3 Fleet Command ←── E3.2, fleet scale

Critical Path

The fastest route to first customer deployment:

E0.1 DMCG Pipeline + E0.3 Observer Wiring (parallel)
E0.4 Curator Pipeline (parallel with above — no dependencies)
E1.1 Customer Onboarding (depends on E0.1 + E0.3)
E1.2 Inefficiency Invoice + E1.3 Auto-Generated Skills (parallel, depend on E0.1 + E0.4)

Everything else is valuable but not on the critical path to first revenue.

Sokrates Implementation Status — what’s shipped today
Technical Architecture Whitepaper — the full vision these epics implement
Product Description and Feature Set — product-level feature spec
Revenue Model and Pricing — the business model these features support
Go-to-Market Strategy — the GTM motion these features enable

Sókrates Wiki

Explorer

Sokrates Roadmap

Sokrates Roadmap

How to Read This

Phase 0: Foundational Infrastructure

E0.1: DMCG Schema Ingestion Pipeline

E0.2: Schema Healing (Hot-Swap)

E0.3: Eidos Observer Wiring

E0.4: Curator End-to-End Pipeline

Phase 1: Product Launch Readiness

E1.1: Customer Onboarding Pipeline

E1.2: Inefficiency Invoice (Monthly Value Report)

E1.3: Auto-Generated Claude Skills from Operational Topology

E1.4: Hardware Tiering (Standard / Sovereign)

Phase 2: Intelligence Layer

E2.1: Gemma 4 Semantic Enrichment

E2.2: Local Inference Stack (DGX Spark)

E2.3: Hypergraph Metalayer (MVP)

E2.4: GQL Query Builder

Phase 3: Scale and Fleet Intelligence

E3.1: Self-Evolution Harness

E3.2: Cross-Fleet Basis Learning

E3.3: Fleet Command (DGX Station GB300)

Research Track

Dependency Graph

Critical Path

Graph View

Table of Contents

Backlinks

Sókrates Wiki

Explorer

Sokrates Roadmap

Sokrates Roadmap

How to Read This

Phase 0: Foundational Infrastructure

E0.1: DMCG Schema Ingestion Pipeline

E0.2: Schema Healing (Hot-Swap)

E0.3: Eidos Observer Wiring

E0.4: Curator End-to-End Pipeline

Phase 1: Product Launch Readiness

E1.1: Customer Onboarding Pipeline

E1.2: Inefficiency Invoice (Monthly Value Report)

E1.3: Auto-Generated Claude Skills from Operational Topology

E1.4: Hardware Tiering (Standard / Sovereign)

Phase 2: Intelligence Layer

E2.1: Gemma 4 Semantic Enrichment

E2.2: Local Inference Stack (DGX Spark)

E2.3: Hypergraph Metalayer (MVP)

E2.4: GQL Query Builder

Phase 3: Scale and Fleet Intelligence

E3.1: Self-Evolution Harness

E3.2: Cross-Fleet Basis Learning

E3.3: Fleet Command (DGX Station GB300)

Research Track

Dependency Graph

Critical Path

Related

Graph View

Table of Contents

Backlinks