These projects show the systems, controls, and evaluation work behind the practice. They are not client testimonials, production guarantees, or evidence of customer adoption.
Open BlueDot IT on GitHubLocal-first, single-user AI assistant with inspectable memory, approved tools, web research, and local activity history.
Demonstrates: Agent product engineering with explicit approvals, inspectable state, and visible tool activity.
Operational safeguards: Cloud providers and websites can still receive prompts or normal web traffic; local-first does not mean universally private.
TypeScript · Node.js · pnpm · Ollama
Inspect repositoryRust-native agent runtime for controlled purple-team and defensive operations with policy gates, constrained WASM execution, memory, and tamper-evident evidence.
Demonstrates: Security boundaries, approvals, evidence, and operational controls integrated into an agent runtime.
Operational safeguards: Controlled execution and WASM constraints reduce authority but do not make arbitrary code risk-free; use remains authorized and supervised.
Rust · Wasmtime/WASI · PostgreSQL · pgvector
Inspect repositoryOpenAI-compatible loopback routing layer for OpenClaw that selects model tiers using embeddings, KNN, and capability gates.
Demonstrates: Observable model routing, capability-aware decisions, bounded failover, and safe handling for tool-driven automation.
Operational safeguards: Provider and capability support is bounded by the implemented adapters; the project does not claim universal model or tool compatibility.
Python · FastAPI · llama.cpp · OpenAI-compatible API
Inspect repositoryOperator-supervised security assessment runtime for authorized engagements with deterministic scope, risk, approval, evidence, and reporting controls.
Demonstrates: Security workflow orchestration where model requests remain inside deterministic operator and evidence controls.
Operational safeguards: Use is limited to authorized engagements; the repository notes that its first public PyPI release is not yet available.
Python · SQLite · CLI · OpenAI-compatible models
Inspect repositoryExploit Intelligence Engine that ingests, normalizes, and scores vulnerability records and exploit-candidate evidence.
Demonstrates: Security data ingestion, evidence normalization, scoring, and remediation-oriented reporting.
Operational safeguards: It scores candidate evidence; it is not an exploit executor or proof of real-world exploitability.
Python · NVD API · SQLAlchemy · FastAPI
Inspect repositoryReproducible benchmark harness for command-line AI agents using isolated workspaces, deterministic assertions, and outcome-verified reports.
Demonstrates: Evidence-based evaluation, reproducible fixtures, and explicit separation of capability coverage from verified outcomes.
Operational safeguards: The harness is not a security sandbox, and one successful trial is not statistically strong evidence by itself.
TypeScript · Node.js · pnpm · JSON assertions
Inspect repositoryBring the application, workflow, or security concern that needs a useful next move.
Request a scoped review