Skip to content

Releases: ThinkFlowLab/system1-agents

Release list

system1-agents 0.1.0

system1-agents 0.1.0 Latest
Latest

Choose a tag to compare

View all tags
@Yunaik Yunaik released this 22 Sep 16:48
v0.1.0
c79697f

First release. Nine agents on TypeSafe Jev, Laya or Cua-S1 Nano, through the CLI and the MCP server, with seven side-by-side replays against a chat model.

Install

git clone https://github.com/ThinkFlowLab/system1-agents && cd system1-agents
uv sync && cp .env.example .env

From Claude Code, Codex or Cursor: npx skills add ThinkFlowLab/system1-agents. The Claude Code plugin: claude plugin marketplace add ThinkFlowLab/system1-agents && claude plugin install s1a@system1-agents.

What ships

  • Nine agents through the CLI and the MCP server: allrecipes and flights (browser use), desktop (computer use on Windows or macOS), ticket_router (30 labelled tickets to five queues), blackjack, game2048, millionaire, alfworld (games and embodied text), injection_guard (rail).
  • Three decision models in the slot: TypeSafe Jev over HTTP, Laya and Cua-S1 Nano in process, plus the llm, random and rule comparison slots.
  • s1a decide and s1a probe for one decision or a JSONL of decisions outside any agent.
  • The caller skill for Claude Code, Codex, Cursor and Hermes, the Claude Code plugin with the s1a-browser subagent, and the builder skill build-s1a-agent.
  • Seven side-by-side replays under docs/assets/demos/, showcase GIFs per eval under docs/results/, and the Google Flights driver comparison in docs/benchmarks.md.

CHANGELOG.md has the full list.

Assets 2
Loading