Dushyant Garg

I build 0 → 1 fast, then scale 1 → 100 on customer data. Open-source tools for AI agents. Writing at dushyantg.substack.com.

~/dushyant-garg $ git log --last-6 LIVE
typed by Jarvis · Friday · Codex · Claude·directed by Dushyant
159f22a5m ago
◉ Claude · directed · Dushyant
feat(seo): add search discoverability foundation (#1)
ec22ae944m ago
◉ Agent · directed · Dushyant
chore(openspec): add SEO discovery change
0500cf61h ago
◉ Agent · directed · Dushyant
feat(observability): add Vercel analytics and speed insights
368f8ae15h ago
◉ Agent · directed · Dushyant
release: bump version to 0.2.1
0f40fbb15h ago
◉ Agent · directed · Dushyant
chore: ignore brainstorming artifacts
37a755615h ago
◉ Agent · directed · Dushyant
feat(lobster): hang around longer — 5.5s message display, slower stride, mid-walk pause

Tools

Memory

Durable state for agents — what they carry across turns and sessions.

AgentSense typescript · sqlite · fts5

Knowledge-graph memory plugin for OpenClaw — entity extraction plus relationship-aware context. Built for agents that need to remember last week.

$npm install agentsense
agentrem typescript · mcp · npm

Structured reminders CLI + MCP server for AI agents — time, keyword, condition triggers with priority escalation, recurrence, and token budgets.

$npm install -g agentrem
github ↗ ·↓ 172 / mo

Libraries

SDKs and primitives for building agents.

AgentTK typescript · cli

Agent-facing CLI primitives. Typed, discoverable, safe by construction — the argv you'd actually hand to a model.

$npm install agenttk
Starglass typescript · mcp · skill

Observation and dispatch layer for agent systems. Stdout, events, tool-calls — one timeline, zero plumbing.

$npm install starglass

Orchestration

Running and coordinating multiple agents.

Potato v3 rust · ratatui · mcp

Personal Orchestration Tool for Agentic Task Operations. Terminal cockpit for coding agents — run Claude, Codex, or others in real PTYs with roles and MCP coordination.

$cargo install potato-v3
agentplan python · sqlite · pypi

Shared to-do list for AI agents. Zero deps, pure Python. Dependencies, priorities, atomic claims — built after three Claudes duplicated a PR on my machine.

$pip install agentplan
github ↗ ·↓ 209 / mo

Human Tools

GUIs, evals, and dev tooling for people who operate agents.

ClawK swift · swiftui · macOS

Native macOS menu bar for OpenClaw. Monitor sessions, heartbeats, memory, and cron jobs from your menu bar — background runs observable without opening a window.

$brew install fraction12/tap/clawk
explain typescript · ts-morph · gh action

Turn any TypeScript/JavaScript codebase into a living, human-readable architecture document. Push to main → GitHub Action runs → your entire codebase is explained in plain English. Built for PMs and founders who build with AI but need to understand what got built.

$npx fraction12/explain
OpenRank astro · supabase · ts

The open benchmark for AI agents — daily puzzles, public rankings, server-side timing. Because "did it work" is not "did it beat the others."

$# visit open-rank.com
WireFlow typescript · canvas

Wireframing and planning app — imagine ExcaliDraw but better. Quick diagrams that stay legible when the meeting moves on.

$# clone + pnpm dev — deploy coming
Agent Office godot · gdscript · openclaw

Watch agents live — not logs, not dashboards, but actually see them moving through a room. OpenClaw-native spatial interface, built in Godot.

$# private preview — reach out if interested
🔒 private · in dev

Products

Customer-facing SaaS. Shipped to real users.

TradeSpec AI python · saas · alpha

AI workflow for trade estimates — faster, safer, grounded in messy contractor source inputs. In alpha with a real customer. Python service (LightningITB) + HTML frontend.

$# in alpha with a customer — reach out for a demo
🔒 alpha · 1 customer live

Experiments

Things I built to learn something. Some worked, some didn't. The lessons travel.

eat next.js · supabase · gpt-4 vision

Kitchen inventory + recipe discovery. AI receipt scanning, 15 RSS feeds, inventory-aware recipe matching. My first Dark Factory build — I never looked at a line of the code. Agents wrote, reviewed, shipped.

$# visit eat-tawny.vercel.app
Vault Mind go · local llm · cli

CLI that talks to your Second Brain through a fine-tuned local model. Experiment parked — fine-tuning at scale was the wrong unit of work for the problem. The learnings are shipping as essays.

$# private experiment — see essays for the postmortem
🔒 parked · learnings published

Writing

All writing on Substack →

Log

Jarvis — my long-running agent — keeps a blog. These are his dispatches from inside the stack.

All dispatches →