Cut AI context cost without trusting the compressor. Every reduction is reversible, byte-exact recoverable, and carries an auditable receipt. Local-first, works through proxy, MCP, SDK, or agent wrapper.
- ai-agents
- claude-code
- mcp
- token-optimization
96,838 visible repositories in the index
145–168 of 188 · Most stars
Cut AI context cost without trusting the compressor. Every reduction is reversible, byte-exact recoverable, and carries an auditable receipt. Local-first, works through proxy, MCP, SDK, or agent wrapper.
:sunrise: A comprehensive, cutting-edge open-source framework for next-generation Transformers/LLM, with compute-matched benchmarking and a platform for autonomous architecture research.
An agent that compiles git maintenance chores into reusable programs, then decides whether to replay them by running checks rather than by guessing the task looks familiar
aiomisc - miscellaneous utils for asyncio
TranslatedClaude Code plugin: turn one still image into a ~5 s seamless looping animation (MP4, GIF, WebP, Live Photo).
Free global SEO/GEO skill for Claude Code, Codex & 10 agent hosts — 18 market portals, 56 stdlib scripts, 672 golden tests, self-update intel loop, monitoring daemon, CI gates. Multilingual GEO, pSEO, CWV, penalty recovery.
Translated☕ Using AI to assist Java / Spring Boot development (AI for Java / AI for Spring) | Requirements, implementation, testing, review, and delivery | Graph → Loop → Workflow → Agent / Skill, fictional runnable examples, and worktree / Docker tools | Aimed at business development, not an implementation of the Spring AI framework
Self-improving code for any git repo: give it a measurable goal, a coding agent (Claude Code or Aider) proposes changes, the kernel scores them, keeps winners and rolls back failures. Open-source AlphaEvolve / FunSearch-style evolution loop, sandboxed, with an auditable ledger.
Solv3R: 3D Looped Transformer as a Multi-view Solver
Open-source AI tutor for personalized learning: Socratic, adaptive lessons to learn anything. Claude Code skill + self-hosted web app. Bloom 2 Sigma · 开源 AI 家教
P&ID to Knowledge Graph: a Claude skill that turns P&ID PDFs (digital or scanned) into a process knowledge graph of flow, instruments and control loops.
Autonomous OODA-loop pentesting agent. Hermes brain, Kali + Playwright tools over MCP, ATT&CK as the Orient language.
Looking Ahead: Modeling Future Evolution in Autoregressive Video Generation
An Agent Skill that writes original instrumental music as editable notes, then performs, mixes and masters it on free sampled instruments (CC0 or free for any use) and synthesizers. MIDI, stems and game loops; no audio-generation model.
Hardware testing for the software world. Real or virtual, local or remote, human, automated or agentic.
Research tooling for MW4 on-demand texture streaming: loopback proxy, hosts manager, resumable mirror, control GUI. Code and notes only - no game files.
Self-OSINT scanner for Windows: look yourself up the way a stranger would, see every public fact about you, then close what you find.
LoopWAM: Closed-Loop Generalist Policy Improvement from Deployment Rollouts
AI trading agents for US stocks with news backtesting that can't see the future: a public LLM trading agent leaderboard (rta-bench-1), frozen point-in-time 2020/2022/2026 markets (prices, VIX, news), a look-ahead check on every run, LLM agents over MCP, and paper/live trading on 10 brokers. 风险感知美股交易智能体 · 新闻回测 · LLM 榜单
Personal project: where shoppers look, estimated from an ordinary security camera and open source AI (YOLO11, Gaze-LLE, Depth Anything V2)
a re-implementation of the asyncio mainloop on top of Trio
Translated知策 (Zhice, "know strategy") | AI investment research and strategy experiment workbench: view markets and news, compare strategy judgments; organize investment documents into knowledge backed by source-text evidence, form rules through manual confirmation, screen candidates, and verify them in independent simulated accounts.
PostgreSQL replication monitoring and failover daemon
Probabilistic programming framework that facilitates objective model selection for time-varying parameter models.