Open Source¶
I have been involved in open source for most of my career. The work spans three GitHub organizations and ranges from production IAM systems to AI knowledge infrastructure.
eXOReaction¶
Synthesis — Knowledge Infrastructure Platform¶
Repository: github.com/exoreaction/Synthesis Language: Java 21 | License: Apache 2.0 | Current version: v1.42.0
Local-first knowledge infrastructure for AI-augmented development. Indexes workspaces (code, docs, PDFs, media, Notion), builds multi-layer knowledge graphs, and exposes everything through a CLI, MCP server, and LSP server.
Key stats: 76 CLI commands · 52 MCP tools · 4,700+ tests · 200--300 files/second indexing
Capabilities:
- Sub-second full-text and semantic search across entire workspaces
- Code knowledge graph (dependency tracking, cross-repo relationships)
- Episodic memory via session indexing (Claude Code session transcripts → searchable SQLite), with remember/recall/reflect MCP tools
- Agent dispatch planner (synthesis dispatch) — pre-populates agent spawn prompts
- Full-stack KCP v0.25 support: manifest scaffolding (kcp init), refresh, verification against evidence (kcp verify), cross-repo federation, read planning (kcp plan), and Ed25519 signing
- Security analysis (code-graph security) — 21 signals across traditional and agentic attack surfaces
- Notion workspace integration with health signals (W022/W023/W024)
- Executive reporting, research engine, maintain/validate/health commands
Release history · Knowledge Infrastructure
xorcery-alchemy¶
Repository: github.com/exoreaction/xorcery-alchemy Language: Java | License: Apache 2.0
Experimental extensions to the Xorcery framework, exploring temporal analytics and DevSecOps intelligence capabilities.
Cantara¶
Cantara builds open-source infrastructure for Java applications: authentication systems, reactive frameworks, messaging abstractions, microservice tooling, and AI knowledge protocols. Most projects are Apache 2.0 licensed and production-ready.
Knowledge Context Protocol (KCP)¶
Repository: github.com/Cantara/knowledge-context-protocol Spec: cantara.github.io/knowledge-context-protocol License: Apache 2.0 | Status: v0.25.1 — submitted to the Linux Foundation Agentic AI Foundation
A YAML file format specification that makes knowledge navigable — and trustworthy — for AI agents. KCP is to knowledge what MCP is to tools: it adds topology (depends_on, supersedes), intent (what question each unit answers), freshness (validated dates), audience targeting, and context window hints. Twenty-five releases in six months have grown it from a table of contents into a full protocol, with layers for discovery, meaning, trust, time, identity, federation, and economics.
RFCs: Auth & Delegation · Federation (Org-Hub) · Trust & Compliance · Unit Content Integrity · Temporal Validity · Payment & Rate Limits (economic metadata) · Context Window Hints
Reference implementations: parsers in Python and Java · MCP bridge servers in TypeScript, Python, and Java · the kcp-agent reference agent (below)
kcp-commands¶
Repository: github.com/Cantara/kcp-commands Language: Java (daemon) + shell | License: Apache 2.0 | Current version: v0.12.0
A Claude Code hook that applies KCP at the Bash tool boundary. Intercepts every Bash tool call: injects concise flag/syntax guidance before execution (no --help round-trips), strips noise after. Also writes every tool call to ~/.kcp/events.jsonl for kcp-memory ingestion.
Measured saving: 67,352 tokens per session — 33.7% of a 200K context window recovered
283+ bundled manifests covering git, Linux, Docker, Kubernetes, cloud CLIs, build tools, package managers, and more. Unknown commands auto-generate manifests from --help.
curl -fsSL https://raw.githubusercontent.com/Cantara/kcp-commands/main/bin/install.sh | bash -s -- --java
kcp-memory¶
Repository: github.com/Cantara/kcp-memory Language: Java | License: Apache 2.0 | Current version: v0.32.0
A Java daemon that indexes Claude Code session transcripts and kcp-commands tool events into a local SQLite database with FTS5 full-text search. The episodic memory layer for Claude Code — session history searchable in milliseconds. Ships as both a CLI tool and an MCP server.
MCP tools: kcp_memory_search · kcp_memory_events_search · kcp_memory_list · kcp_memory_stats · kcp_memory_session_detail · kcp_memory_project_context
kcp-agent¶
Repository: github.com/Cantara/kcp-agent Language: TypeScript (npm) + native binaries | License: Apache 2.0 | Current version: v0.19.1
The reference agent for KCP — a deterministic, fail-closed navigation planner plus optional LLM synthesis. It reads a knowledge.yaml, scores and gates units against declared trust, freshness, audience, and budget, and produces an inspectable plan before any content is loaded or any model is called. Determinism at the core, the model at the edge. Since v0.11 the surface has grown into the full gate triad — load (plan), assert (ground), act (assess) — plus per-unit decision traces (--trace), versioned plan JSON with diff and replay, episodic memory (remember/recall), serving-endpoint discovery, and 13 published conformance vectors other implementations can test against. Ships as an MCP server (kcp_plan, kcp_load, kcp_validate) and as self-contained native binaries.
npx kcp-agent plan "your task" --manifest https://example.com/knowledge.yaml
claude mcp add kcp -- npx -y kcp-agent mcp
See the Defendable Agents field guide for the full architecture.
kcp-harness¶
Repository: github.com/Cantara/kcp-harness Language: TypeScript | License: Apache 2.0 | Current version: v0.10.3
Deterministic knowledge governance for any AI agent. An MCP compliance proxy that sits between an agent and its tools, routing every knowledge request through a deterministic multi-gate governance cascade — fail-closed policy, append-only audit trail, budget ledger, and temporal pinning — and emitting the decision traces, audit logs, and budget records that map onto SOC 2 / ISO 27001 / GDPR controls. No model involvement in the governed decision. Recent releases extend governance from knowledge reads to actions: human approval workflows for held requests, temporal watch (re-decide a standing plan when time alone changes the outcome), action conformance checking, resolution signatures, and an x402 payment wallet under hard budget ceilings.
kcp-skill¶
Repository: github.com/Cantara/kcp-skill Language: TypeScript/Node | License: Apache 2.0 | Current version: v0.1.0
Conventions, linter, conformance vectors, and a curated library for governed skills — kind: skill units whose declared action_scope (tools / paths / capabilities, KCP v0.26 §4.3a) bounds what an agent may touch when enacting a procedure. The repo owns the layer between the spec and the implementations: an authoring profile ("author the envelope like a firewall rule — start from nothing"), 8 lint rules (SK001–SK008, fail-closed exit codes), 5 conformance vectors, and starter playbooks (key rotation, verified dependency bumps, release-and-tag) that any KCP-aware harness can adopt. Enforcement stays with pi-kcp / kcp-harness; evidence-based generation is tracked in Synthesis #477.
pi-kcp¶
Repository: github.com/Cantara/pi-kcp Language: TypeScript | License: Apache 2.0 | Current version: v0.3.2
KCP agent proficiency and ergonomics for the Pi coding agent harness. Human-facing /kcp commands (plan, validate, recall, health) and agent-facing skills on stock Pi — no MCP client required — with kcp-agent invoked as a CLI for deterministic plans and kcp-memory over HTTP for episodic recall. Its runtime-depth milestone added enforcement: a skill declared in knowledge.yaml with an action_scope becomes enforced authority — native tool calls outside the declared tools/paths/capabilities are blocked by the same deterministic decision the kcp-harness proxy makes, and every governed turn carries one correlation id. Code intelligence (Synthesis or any other provider) stays behind Pi's MCP configuration as an optional, substitutable backend.
Companion post: The AI Agent That Keeps the Receipts · interactive demos
kcp-dashboard¶
Repository: github.com/Cantara/kcp-dashboard Language: Go | License: Apache 2.0 | Current version: v0.27.0
A live terminal dashboard (Bubble Tea + Lip Gloss) for the KCP ecosystem — guidance effects, session profiles, token savings, and memory-search activity, auto-refreshing every two seconds.
Whydah — SSO / IAM Platform¶
Repository: github.com/Cantara/Whydah (16 repositories) Website: getwhydah.com Language: Java | License: Apache 2.0
A complete Single Sign-On and Identity & Access Management solution. Whydah handles user authentication, token management, role-based access control, and user administration. It is production-deployed across multiple Norwegian enterprise clients.
Core modules:
- Whydah-SecurityTokenService — Application and user token management
- Whydah-UserIdentityBackend — Identity storage with LDAP integration
- Whydah-SSOLoginWebApp — SSO login interface
- Whydah-UserAdminService — User administration backend
- Whydah-UserAdminWebApp — Admin UI
- Whydah-Java-SDK — Java integration SDK
Quick start (Docker):
docker pull whydah/whydah-all-in-one-image
docker run -it -p 9997:9997 whydah/whydah-all-in-one-image
# http://localhost:9997/sso/welcome (admin / whydahadmin)
Xorcery — Reactive Java Framework¶
Repository: github.com/Cantara/xorcery Language: Java 21+ | License: Apache 2.0
A modular Java library framework built around dependency injection (HK2), composable YAML/JSON configuration, and reactive streams. Designed for building highly performing microservices with strong operational characteristics out of the box.
Key features: - Composable configuration (YAML + JSON Schema) - Reactive streams over WebSockets - OpenTelemetry integration - 30+ extensions: AWS, certificates, DNS, EventStore, JAX-RS/Jersey, Jetty, JWT, OpenSearch, and more - Used as the foundation for the Xorcery AAA (Alchemy + Aurora) analytics platform
Stingray — Microservice Application Framework¶
Repository: github.com/Cantara/stingray Language: Java | License: Apache 2.0
A Java application framework with strong conventions for building microservices. Provides structure, configuration, and lifecycle management. Used as the base framework in large-scale deployments (34+ services in production).
Messi — Messaging Abstraction¶
Repositories: MessiSDK + provider libraries Language: Java | License: Apache 2.0
A messaging and streaming abstraction layer with pluggable providers for different backends. Write once, switch providers without code changes.
Providers:
- MessiS3Provider — AWS S3
- MessiSQSProvider — AWS SQS
- MessiKinesisProvider — AWS Kinesis
lib-electronic-components¶
Repository: github.com/Cantara/lib-electronic-components Language: Java | License: Apache 2.0
A Java library for working with electronic components in manufacturing contexts. Provides MPN normalization, component similarity analysis, BOM management, and manufacturer data for 135+ manufacturers.
Capabilities: - MPN normalization and type detection - 17 specialized similarity calculators (resistors, capacitors, MOSFETs, MCUs, sensors, and more) - BOM (Bill of Materials) creation and validation - Alternative component finding - Component categorization with metadata-driven profiles
Nerthus / Visuale — Service Visualization¶
Repositories: nerthus · nerthus2 · visuale Language: Go / Java | License: Apache 2.0
Real-time dashboards for visualizing microservice environments. Shows service health, deployment status, and version distribution across a fleet of running services.
Infrastructure & Utilities¶
microservice-baseline — Starting point template for building well-structured microservices.
HTTPLoadTest-Baseline — Load testing tool designed for integration into CD/CD pipelines.
ConfigService — Centralized configuration management service with SDK and dashboard.
Valuereporter — Metrics and analytics collection: observations, activities, and reporting.
property-config — Lightweight property-based configuration management.
realestate-metasys-cloudconnector-agent — Reads sensor data from Johnson Controls Metasys building automation systems and distributes to cloud.