Be: pretty good at computers, thinking machines & systems.
Climbing the k-tech tree with multi-agent AI systems. Proud owner of 10x Agent Cluster.
Deep in agent harness, workflows & tools at ludicrous speed. AS of 2026, focused on RSI, kernels, RL infrastructure, self-improving and personalized AI systems, distributed systems, agentic coding and more.
- 🧠 Mem-RLM - memory augmented inference library for Recursive Language Models (RLMs) | Open-Source
- 💾 Agent Skills for Compute - self-improving skill catalog for AI agents — 21 skills covering the full LLM lifecycle, autonomous research, GPU/TPU/QPU kernels, and scientific computing | Open-Source
- 📄 ContextJira - AI-native context extraction from Jira | Open-Source
- Mem-RLM — Memory-Augmented Inference for Recursive Language Models
- ContextJira: AI-Native Context Extraction from Jira
- Claude Code-Time Skill Acquisition with Agent Teams
- A Self-Improving Skill Catalog for AI Agents
- On Compression, Computation and the Space Between
- What coding agents I am using? Codex 5.4 XHigh + Opus 4.6 + Gemini 3.1 Pro at a time.
Weekly Tokens Consumption
| Date | Input | Output | Cache Create | Cache Read | Total Tokens |
|---|---|---|---|---|---|
| 2026-03-10 | 266,170 | 309,700 | 49,037,740 | 1,352,638,440 | 1,402,252,050 |
| 2026-03-09 | 401,990 | 196,460 | 39,007,660 | 1,407,421,650 | 1,447,027,760 |
| 2026-03-08 | 69,350 | 280,760 | 36,715,490 | 1,045,437,200 | 1,082,502,800 |
| 2026-03-07 | 476,630 | 313,070 | 27,446,520 | 1,063,478,780 | 1,091,715,000 |
| 2026-03-06 | 294,340 | 348,320 | 26,253,320 | 936,220,950 | 963,116,930 |
| 2026-03-05 | 60,760 | 122,180 | 18,007,440 | 430,776,410 | 448,966,790 |
| 2026-03-04 | 269,390 | 226,140 | 32,562,510 | 718,528,430 | 751,586,470 |
| 2026-03-03 | 332,080 | 205,780 | 19,330,680 | 519,126,370 | 538,994,910 |
| 2026-03-02 | 27,642 | 66,507 | 4,029,774 | 100,036,680 | 104,160,603 |
| 2026-03-01 | 276,420 | 665,070 | 40,297,740 | 1,000,366,800 | 1,041,606,030 |