Kevin Ruschman
I build local-first agent infrastructure so coding agents and company knowledge systems can plan and test without writing your disk until you approve — state in Postgres, models interchangeable.
Language models are probabilistic samplers; modifying your codebase or managing confidential client knowledge requires a deterministic transaction manager. My open-source architectures replace hallucination-prone prompt chains with hard transactional invariants, sandboxed test verification, and atomic two-phase commit disk updates.
Creator of krusch (coding harness), kd-Code (developer workbench with center-stage diff review), krusch-context-mcp (a 71★ sovereign 16-tool Model Context Protocol server), and krusch-cascade-router (sub-15µs CPU syntactic gates with neural centroid escalation).
The Invariant Lifecycle: Stage → Verify → Commit
How krusch and kd-Code prevent coding agents from corrupting git repositories or applying broken refactors.
The model plans and generates diffs directly into relational tables (staged_diffs). Zero unvetted file writes ever touch your physical working tree.
A shadow workspace runs automated test suites, typechecks, and linters in isolation. If a test fails, the staged turn aborts instantly without repo pollution.
Inspect side-by-side color-coded diffs via @pierre/diffs, symbol graph dependencies, and AST impact analysis before authorizing disk modification.
Only on verified tests AND explicit human approval does the two-phase commit manager fsync changes to disk. Instant 1-click rollback preserves sanity.
Explore by Need: Three Doors
Select your entry point across developer harnesses, persistent memory, and sovereign enterprise pilots.
Coding Harness & Workbench
Stop coding agents from hallucinating broken repositories. Models plan in Postgres, test in sandboxes, and apply changes strictly via atomic 2PC commits.
Memory & Cascade Routing
Eliminate agent amnesia and slash routing costs. Unified 16-tool Model Context Protocol server featuring temporal decay and AST graphs, paired with sub-15µs CPU routing.
Sovereign On-Prem Systems
Air-gapped document intelligence and high-throughput data engineering for regulated practices and high-stakes operations. Deterministic page-and-line evidence grounding, messy OCR pipeline confidence, and de-risked 30-day pilots.
Core Architectural Invariants
The design rules governing every system in the KruschDev ecosystem.
Invariants Over Prompts
Never trust language models to police their own file writes or follow conversational rules. Enforce constraints in PostgreSQL ACID tables, AST parsers, and atomic two-phase commit journals.
Multi-Tier Routing Cascades
Don't burn a 500ms LLM call just to choose an LLM. Intercept code fences and SQL on CPU in <15µs for $0.00, escalate to sub-50ms neural centroids, and only query frontier models when necessary.
Local-First Sovereignty
Run compute, vector indexes, and episodic memory on self-hosted bare-metal clusters. Privileged files, client discovery, and organizational memory never leak to third-party cloud APIs without explicit consent.
Technical Insights & Publications
Architectural breakdowns on memory substrates, agent steering, and sovereign infrastructure.
⚖️ The Sovereign Triad: Unifying Document Ingestion, Statutory Graphs, and Agentic Working Memory
Why autonomous AI agents fail in complex legal and enterprise domains. Systems architecture deep dive into how KruschNexus (layout-true ingestion), KruschLaw (statutory authority & assertion grounding), and KruschContext MCP (Lakebase working memory & trajectory auditing) create a closed-loop, self-auditing intelligence substrate.
Read Sovereign Triad Architecture Essay →⚡ The Ingestion Imperative: Why Retrieval Quality Is Bounded Before the Model Reads a Single Word
Why wrong retrieval is the dominant root cause of enterprise AI failure. Systems architecture deep dive into embedding space geometry (measured negation and numeric limits), parser contracts, citation spines, and how the structured fix itself fails.
Read Systems Architecture Note →⚖️ Citations Die in Local RAG: Why We Built KruschNexus and What It's For
Why arbitrary token slicing destroys document citations. Systems architecture deep dive into KruschNexus (v0.2.3): span-accurate PDF bounding boxes, Poppler TSV multi-column ordering, fail-closed hybrid retrieval, ungameable multi-metric evaluation across legal instrument families, and air-gapped homelab resilience.
Read KruschNexus Architecture Deep Dive →⚖️ Why We Built KruschLaw: Local-First Legal RAG & Ordinance Intelligence
Why generalist RAG fails in statutory law and municipal codes. Systems architecture deep dive into KruschLaw (v0.2.0): hybrid PostgreSQL ranking (tsvector + pgvector RRF), element-aware chunking, automated citation & verbatim quote grounding scanners, and zero-cloud confidentiality for law firms.
Read KruschLaw Architecture Deep Dive →⚡ Multi-Tier Cascade Routing: Slashing Enterprise Inference Budgets by 60–80%
Don't burn a 500ms LLM call just to choose an LLM. Systems architecture deep dive into krusch-pre-router (<15µs CPU gate, $0.00 routing tax), Stage-1 neural centroids (<50ms cosine routing), and speculative parallel hedging to eliminate fallback latency.
Read Cascade Routing Deep Dive →⚡ Sovereign Context & Working-Memory Architecture for Coding Agents
Decoupling tool density, stratified decay, and transactional state for enterprise codebases. Systems architecture deep dive into krusch-context-mcp (v1.6.x) and the krusch (v0.1.0) 2PC apply harness: 2026 Streamable HTTP, domain-stratified λ-decay, declarative steering nuggets, and sovereign PostgreSQL 16 + pgvector persistence.
Read Context Architecture Deep Dive →🛡️ The Krusch Coding Harness: Why Autonomous Code Agents Require a Transaction Manager, Not a Prompt
Why unconstrained file writing breaks coding agents. Systems deep dive into krusch: pre-commit SHA-256 diff staging, Bubblewrap container isolation, storage-engine 2PC apply journals, row-locked catalog FSMs, and actionable failure attribution via Modular RSI.
Read Harness Architecture Deep Dive →⚡ Postgres as the State Plane: Side Effects Require a Transaction Manager
A language model is a probabilistic sampler; side effects require a transaction manager. How krusch (coding harness), krusch-pre-router (sub-15µs CPU gate), krusch-cascade-router, and krusch-context-mcp unify into an invariant-gated developer harness.
Read Full Architecture Article →🧠 Beyond Simple RAG: What Makes krusch-context-mcp Special
Exploring the core capabilities of krusch-context-mcp: mathematical temporal recency decay, declarative steering facts, company brain consensus, lakebase SQLite sync, and research engines.
Read Architecture Deep Dive →⚡ Architecting Code-Agent Retrieval & Memory: Beyond Naive Vector Search
Why sliding-window RAG fails on codebases. Systems deep dive into krusch-context-mcp: single-store PostgreSQL 16 + pgvector, structural symbol extraction, SQL Reciprocal Rank Fusion (RRF), dependency graph walks, and deterministic server token budget packing.
Read Retrieval Architecture Article →🚀 Local-First Core, Hybrid Cloud Elasticity: Decoupling Agent Memory
Decoupling AI coding agent working memory into managed PostgreSQL (pgGraph + HNSW) and optional cloud embeddings for zero VRAM load and global context while keeping file writes strictly sovereign.
Read Full Article →