I build open-source infrastructure for AI agents at @tinyhumansai, mostly in Rust. Writing code on GitHub since 2016. This account and @jainsanil18 are both me, and the card below counts them as one.
| 🐝 OpenCompany | Runs a hive mind of agents toward a single goal. My main project: 86 merged PRs. |
| 🧠 OpenHuman | A fast, low-cost open-source agent harness (42k★). Core runtime and benchmarks. |
| 📊 openhuman-benchmarks | Public evals like Terminal-Bench 2. I find failures caused by the harness, not the model, and fix them. |
| 🦀 tinyagents · tinyflows | A provider-neutral agent harness and agentic workflows, both in Rust. |
| 💾 tinymemory | A common memory interface for agents. I work on recall quality and evals. |
Also: tinytools · tinyartifacts · tinymcp · tinysearch
- 📡 DPSN: real-time data streams for AI agents. I built deepsense, an agentic LLM orchestration framework that acts on live data, plus the Node.js SDK, the SDK docs and the agent UI.
- 🎙️ Talker Network: real-time push-to-talk voice.
2026-11-05tinyhumansai/openhuman · feat(sandbox): configure agent37 box CPU and memory2026-11-05tinyhumansai/openhuman · Fix agent37 sandbox routing, constraints, template, egress, and cleanup2026-11-05tinyhumansai/tinybox · Admit agent37 module Exec through its own HTTP channel2026-11-04tinyhumansai/openhuman · Add the agent37 sandbox backend wiring2026-11-04tinyhumansai/tinybox · Add the agent37 hosted HTTP sandbox backend2026-11-03tinyhumansai/tinysearch · Return inline snippets from web_search and steer callers to use them2026-11-03tinyhumansai/openhuman · fix(shell): resolve the sandbox backend before reserving a process slot (#8317)2026-11-02tinyhumansai/openhuman · fix(voice): classify hosted live-voice billing refusals instead of reporting them
agent harnesses · evals & benchmarks · long-running agents · agent memory · Rust
The activity card and feed are regenerated daily by a GitHub Action from both accounts.