AI Engineer · Full Stack Architect · Founder @ NorahLabs

Building production AI systems from Nairobi, Kenya. Agentic orchestration engines, retrieval architecture, and native desktop applications — designed for developers who want real architectural understanding without cloud lock-in.

The Open-Weight Frontier Harness — the agent workstation built for open-weight frontier labs, where a weaker local model performs like it knows your codebase.

v13.6.0 · macOS / Linux / Windows · 396MB (leaner than Cursor) · atlarix.dev

Atlarix is built for the open-weight frontier labs — DeepSeek, Qwen, Kimi, MiniMax — not just compatible with them. The bet: durable open-weight labs, not specific model versions. What makes a weaker model punch above its weight is Blueprint — a section-scoped structural index (Universal Ctags symbol graph + ast-grep import/call/route edges + a disk-backed SQLite FTS5 lexical layer) that hands the agent ~6,500 tokens of real structural understanding instead of a whole-repo dump. No embeddings, no vector store — structural + lexical retrieval, with the model as the natural-language layer. Any-model BYOK, local-first execution, every destructive action behind your approval.

Published research: Blueprint: Section-Scoped Structural Graph Retrieval and Post-Turn Compression for Agentic LLM Coding in Multi-Repository Workspaces · Zenodo (CERN), 2026

What makes it different:

• Built for open-weight, not retrofitted — DeepSeek / Qwen / Kimi / MiniMax as first-class, plus broad BYOK (OpenAI, Anthropic, Gemini, OpenRouter, Groq, Together, Mistral, xAI, Hugging Face) and local Ollama / LM Studio.
• Blueprint, embedding-free — Universal Ctags + ast-grep edges + SQLite FTS5, reranked by structural relevance (reciprocal-rank fusion, proximity scoring) at constant low memory for any repo size. grep returns the most relevant files first, each annotated with its enclosing symbol.
• Atlarix Core — four managed open-weight models, Core 1 (fastest) → Core 4 (most capable); every plan can use every model, allowance differs by tier. No API key required on the free tier.
• Parallel sub-agents — the task tool fans out up to five concurrent read-only scouts per turn, live thinking streamed per agent.
• Verified edits — in Build / Debug the agent runs the project's own checks (tsc/eslint/ruff/mypy/pytest) through a sandboxed terminal and can't declare a task done while they fail.
• Per-OS security sandbox — write-confining command execution, an approval queue with hunk-level diff accept/reject, committable permission + hook rules at a single execution funnel.
• Token-efficient by design — on-demand tool-driven retrieval and model-triggered compress_context; provider-native tool blocks and prompt caching lift weak / local model performance.
• Real workspace — interactive PTY terminal (persists across restarts), Monaco editor, in-app browser, @ file/folder mentions scoped to the current workspace, per-workspace MCP.

Work modes: Explore (read-only) · Plan · Build · Debug · Review (correctness + security)

Pricing: Solo (free) · Studio ($19/mo) · Team ($79/mo)

Achievements: Winner — Amazon Nova AI Hackathon (Devpost × AWS, 2026) · GDG Nairobi Agentathon (2026) · Lua × Antler (Nairobi, 2026) · Red Bull Basement Global Innovation Programme

Blueprint: Section-Scoped Structural Graph Retrieval and Post-Turn Compression for Agentic LLM Coding in Multi-Repository Workspaces Amariah Kamau, NorahLabs — 2026

Documents the Blueprint architecture, post-turn tool-result summarisation, and benchmark results from a controlled A/B evaluation on a production multi-repository workspace — including a counterintuitive finding: structural confidence lets the agent explore more, using more total context but producing stronger results.

atlarix-skills — Community Agent Behaviors registry. SKILL.md files teaching agents language patterns and framework conventions across React, Next.js, Python, TypeScript, Go, Rust, Docker, MCP, and more. Apache 2.0.

atlarix-mcps — MCP server registry for Atlarix. One-click connections to Gmail, Calendar, Drive, and more from within the agent environment. Apache 2.0.

atlarix-releases — Desktop app release builds for macOS, Linux, and Windows.

Praxia — AI healthcare assistant with MONAI X-ray analysis (pneumonia, fractures, tumours), multilingual symptom diagnosis, real-time WebSocket chat, and a Docker/Nginx/Celery production stack.

Commit Checker — Open source CLI for Git commit analysis.

Atlarix is built in Nairobi. Most serious AI developer tooling assumes a Western cloud subscription. Atlarix is open-weight by design — bring any provider, including locally-built African models. Exploring native MansaLLM integration with African AI labs.