Language · Rust + LLVM · Apache 2.0
hotlang
A very small, purpose-built language for the hot path of a high-frequency trading system.
hotlang covers one thing: the decision kernel, meaning book signals, rolling statistics and strategy math. It compiles to a native object through LLVM and your host program in C, C++, Rust or Java calls it like a C function; feed handlers, order gateways and all I/O stay in the host. What makes it different is what it refuses to have. If a function compiles, the compiler has proven it cannot allocate, cannot loop forever, cannot recurse and cannot index out of bounds. Every operation is defined on every input, and if compiles to a select rather than a branch, so per-call cost does not depend on the data. Because those are exactly the properties a circuit needs, the same source also compiles to Verilog for an FPGA. On an isolated kernel it ties hand-tuned C++, which is physics and the README says so; the win is that the accidental allocation someone adds in month six is unwritable, so the performance floor is held by the compiler instead of by vigilance.
github.com/vikasgrac/hotlang →
Claude Code plugin · Python · MIT
project-brain
Every Claude Code session, in every project, captured into one plain-Markdown brain you own.
Claude Code forgets: transcripts are deleted after 30 days and memory is scoped per project and per machine. project-brain fixes that with the simplest thing that works, Markdown files in a folder you control. When a session ends it is archived as readable Markdown with API keys, tokens and private text stripped, distilled into decision, fact and open-thread notes, and compiled into a living page per project. When a session starts, only a small index is injected; the notes load on demand, so the token cost stays flat and there is no vector database, no service and no subscription. If you uninstall it tomorrow, your knowledge is still plain Markdown.
/plugin marketplace add vikasgrac/project-brain then /plugin install project-brain@project-brain.
github.com/vikasgrac/project-brain → · Read the write-up
Terminal tool · Python (Textual) · MIT
claude-reader
A live reading pane for Claude Code sessions: only the text Claude writes to you, in any terminal.
In a long Claude Code session the prose that matters to you is buried in tool calls and command output, and the terminal never renders the markdown. claude-reader tails the transcript file Claude Code already writes, keeps only the assistant text and your own prompts, and renders them in a sidebar plus reading pane that updates twice a second. No hooks into Claude Code, no plugin, no server: it only reads a file, so it works in any terminal split, over ssh, and does not break when Claude Code's UI changes. Independent project, not affiliated with Anthropic.
pipx install claude-reader or uvx claude-reader.
github.com/vikasgrac/claude-reader → · PyPI · Read the write-up