agentbelt 0.3.0: token optimization for local AI agents, with honest numbers

The developer behind agentbelt released version 0.3.0, adding a Goose adapter and demo GIF to the MCP-based tool that gives AI coding apps like Claude Code, Codex, Cursor, Copilot, DeepSeek and LM Studio shared, audited dev tools. Benchmarking context compaction on a local gpt-oss 20B model with a 16k window, the developer found compaction cut average input tokens on a repeated failing-test-log task from 42,267 to 18,448 while leaving accuracy unchanged at 1/3, but slightly increased tokens on a five-file read task (46,916 to 50,410). The stated takeaway is that avoiding repeated logs is the biggest win for local agents, while shrinking file contents can backfire.

agentbelt https://github.com/LovaRK/agentbelt gives every AI coding app Claude Code, Codex, Cursor, Copilot, DeepSeek, LM Studio the same safe, audited dev tools over MCP. Version 0.2.0 added context compaction for agents running on local models with small context windows gpt-oss 20B, 16k context, on a MacBook Air . 0.3.0 adds a Goose adapter and a demo GIF. I ran the same tasks with compaction off and on, 3 runs each. Token counts are what Ollama itself reported. | Task | Compaction | Correct | Avg input tokens | Largest prompt | |---|---|---|---|---| | read 5 files | off | 3/3 | 46,916 | 13,317 | | read 5 files | on | 3/3 | 50,410 | 10,269 | | failing test log, twice | off | 1/3 | 42,267 | 6,182 | | failing test log, twice | on | 1/3 | 18,448 | 3,338 | Takeaway: for local agents, the biggest win is not repeating the same logs. Shrinking file contents can backfire. Full method and raw results: docs/token-optimization.md https://github.com/LovaRK/agentbelt/blob/main/docs/token-optimization.md npx agentbelt setup Then ask your AI app: "Use agentbelt's agent run with compact: true to …" ⭐ If this is useful, a star on GitHub https://github.com/LovaRK/agentbelt helps other developers find it. And if you run agents on local models, I'd love to hear what you've measured.