Trustworthy systems
for the agent era.
Open-source infrastructure that makes AI agents, cloud development, and software supply chains safe enough for real work.
Work that stands on its own
Distinct builds across agent infrastructure, developer tools, supply-chain security, secure data exchange, and local cloud infrastructure — each with its own brief and craft.
Aletheia
The unconcealment layer for autonomous agents.
Twelve products on one stack — governance, verification, optimization, infrastructure and the physical frontier. Every action an agent takes is constrained, verified, attributed and auditable, so nothing it does can be hidden.
- Layered constraint enforcement — each layer bounds the one below
- Formal verification of agent-generated code against specifications
- Causal provenance graphs with cost and energy attribution for every decision
- Self-healing infrastructure with regression detection and rollback
ax
The AI coding agent built without the trade-offs.
A terminal-native coding agent with local-first architecture and security-first defaults. Multi-provider, cost-aware, and secure by default — own your agent instead of renting it from a single vendor.
- Multi-provider: Anthropic, OpenAI, Gemini, Ollama, vLLM
- Local-first architecture with security-first defaults
- 30+ built-in tools, sub-millisecond code search, 253 tests
EvolveHandsAI
Code that breeds better code.
An evolutionary optimization engine that breeds implementations instead of prompting for them. Define a fitness function and mark the mutable regions; an LLM ensemble evolves hundreds of candidates across island populations, MAP-Elites preserves diversity, and anything that breaks the tests is rejected automatically. The page itself runs a real engine in your browser — every gauge is live.
- LLM ensemble — Claude, GPT-4o and Haiku with adaptive model weighting
- MAP-Elites quality-diversity grid across feature dimensions
- Island parallelism — 2–5 populations with periodic migration
- Isolated Docker sandboxes: read-only filesystem, no network, 256 MB · MIT licensed
VIM
Vulnerability Inheritance Map.
Frontier models now find zero-days faster than anyone can track where they live. VIM is the lineage layer: an open-source graph database that maps a single upstream finding to every fork, vendored copy, and compiled binary that inherits it. Apache 2.0, targeting the OpenSSF Sandbox.
- Traces vulns across forks, vendored deps & binaries
- Graph-native blast-radius analysis
- Apache 2.0 — aimed at the OpenSSF
Wharf
A dock for your Managed Agents.
An MCP server that exposes Anthropic's Managed Agents API through a conversational interface. Deploy, message, update and retire autonomous agents — all from Claude, without leaving the chat.
- MCP server for Anthropic Managed Agents
- Deploy, message, update & retire from chat
- Runs where your Claude client already is
X-Road
Secure data exchange that watches, predicts, and explains itself.
X-Road moves data between organisations with end-to-end integrity and non-repudiation. This edition adds an opt-in intelligence layer that runs beside the system — reading operational metadata only, never a payload, never in the signing path — to detect anomalies, correlate incidents and brief operators in plain language.
- Anomaly detection over federated temporal call-graph models
- Collapses edge alerts into correlated incidents, with breach forecasting and confidence scores
- Grounded copilot — source-cited, plain-language incident briefings
- Read-only sidecar on X-Road 7.8.1 · 0 payloads read · advisory-only · MIT
LoopbackCloud
The local cloud for AI agents.
Runs AWS, Azure and GCP on your own machine so agents — and the people building them — can provision and stress real resources with zero spend and zero blast radius. The same SDKs, CLIs and IaC you already use, behind one unified console with a gated multicloud interconnect.
- 70+ services across AWS, Azure & GCP in one local stack
- Drop-in for standard SDKs, CLIs, Terraform & CDK
- Built-in MCP server so agents can inspect, provision and stress real infrastructure
- Native GraalVM image — 24 ms cold start, MIT licensed
Trust isn't a feature you bolt on at the end. It's the architecture you start with — in your agent, your cloud, and your supply chain.Abdul Jaleel Kavungal — the studio principle
Let's build something worth trusting.
Open to opportunities, collaborations, and design-partner conversations across AI, platform engineering, and supply-chain security.