Projects
Software I build or maintain, grouped by where it sits. Most of it is open source; research code lives on the research page.
Inference and models
Engines, model artifacts and serving infrastructure.
-
memra
tiyuvta's inference engine in Rust and CUDA, tuned for RTX PRO 6000, RTX 5090 and B200. Safetensors is the tuned path and GGUF stays supported. Prebuilt binaries on crates.io.
-
tiyuvta on Hugging Face
The lab's published models: NVFP4 builds of GLM, DeepSeek, MiMo, Qwen and DictaLM, MTP draft heads in GGUF, and Phonon-2 in ONNX.
-
hqmtp
The draft-head research harness behind working paper 1, with the paper draft and every run ledger.
-
sglang-valkey-demo
Valkey as the placement and event plane for SGLang's cache-aware routing, measured through indexer restarts and fleet rotation.
Valkey and data systems
In-memory data, clients, queues and durable execution.
-
Valkey GLIDE
The official multi-language Valkey client: a Rust core with Python, Java, Node.js and Go clients.
-
valkey-skills
The official AI skills for the Valkey ecosystem. I started it.
-
Valkey
Contributor to Valkey itself. Sync-from-replica replication with a delayed primary switch is in review.
-
glide-mq
Node.js queue on Valkey Streams, built on Valkey GLIDE's Rust-native bindings. Budgets, streaming and pauses are part of the job model, with adapters for Hono, Fastify, Hapi and NestJS, and a dashboard.
-
FlowFabric
Durable-execution engine for Valkey, Postgres and SQLite: lease-safe workers, waitpoints and budgets.
-
ocaml-valkey
Valkey client for OCaml 5 and Eio, RESP3 only. Published on opam.
-
CRIU live migration
Copy-on-write live migration of a 200 GB loaded in-memory datastore with under 50 ms of freeze: userfaultfd page prefetch, priority-ordered memory transfer and lazy-page fixes.
Agent tooling
Tools for agent-assisted development, mostly under agent-sh.
-
agnix
Linter and language server for AI agent configuration, with autofixes, a GitHub Action, an MCP server and editor plugins.
-
parlar
Voice mode for Claude Code and Codex: talk to a running session and it answers out loud. Runs locally on CPU. On npm and crates.io.
-
computer-use-linux
Linux desktop control over MCP: accessibility tree, GNOME Shell, Wayland portals.
-
agent-workspace-linux
Isolated, agent-owned Linux desktops and browsers over MCP, so an agent never has to touch your real machine.
-
agentsys
The agent-sh plugin set (workflow, review, ship, deslop, perf) for Claude Code, OpenCode, Codex, Cursor and Kiro.
-
revuto
Local PR reviewer that works with any model and learns each repository from its PR history and maintainer feedback. It reviews my own repos.
-
harness tools
Read, write, grep, glob, bash, webfetch, LSP and skill tools built for LLM callers, published as @agent-sh/harness-* on npm with Rust ports at parity.
-
linubot
Linux desktop app for a team of AI helpers: per-bot providers, memory and isolated computer workspaces.
-
Codex Desktop for Linux
The unofficial Linux build of the ChatGPT and Codex desktop app: Linux computer use, read-aloud and conversation mode, the in-app updater, launcher hardening.
Systems tools
Close to the operating system.
-
ferrings
io_uring TCP transport for Node.js with a Rust N-API core. 2.5x the throughput of Node's http module and 37 to 57% fewer syscalls per connection in the published benchmark. On npm.
-
layout-audit
Memory-layout analysis from DWARF: padding, layout diffs and size budgets for C, C++, Rust and Go. A CLI and a GitHub Action.
-
scrump
Format-aware secret scrubber for binary capture artifacts: perf.data, core dumps, nsys traces, JFR.
-
RustOwl
Visualizes ownership and lifetimes in Rust. I worked on its runtime, memory use and CI.