Trim 1

Trim 1 is a small model that selects and keeps only the necessary lines of a context for an AI agent query.

Trim 1 is a small model designed to help AI agents process long contexts by deciding which lines are needed for a current query and dropping the rest. It takes inputs such as tool results, files, logs, or long chats and returns the original lines in their original order with markers showing where removals occurred. It never rewrites or summarizes lines, ensuring identifiers, paths, and error text remain exact. Built as 712 gradient-boosted trees totaling 1.2 megabytes, it runs in milliseconds on a single CPU core without a GPU or network calls. The software features deterministic execution with no sampling, has no dependencies outside the Go standard library, and is written in pure Go. It includes settings to adjust cutting intensity between conservative, balanced, and aggressive options.

Trim 1 at a glance

Best for: Software engineer
Runs on: CLI
Pricing: Free
Released: October 2026
Category: AI productivity tool
Website: github.com/caveman-ai/trim-1

Key features

Selects relevant lines based on context and query
Maintains original line order with removal markers
Executes in milliseconds on one CPU core
Operates deterministically without sampling or randomness
Runs with pure Go and zero external dependencies
Supports conservative, balanced, and aggressive cut settings

Use it for

Trimming long documents for AI agent retrieval
Processing coding-agent tool outputs
Filtering logs and file contents for relevant data

Alternatives to Trim 1

Basic Memory — Stores data in local Markdown files that remain accessible and readable by any AI.
IWE — Works natively across Mac apps, text editors, terminal CLI, and MCP servers.
Docling — Provides advanced PDF understanding and local execution for sensitive data.
Pro Workflow — It creates persistent, searchable memory and wikis for coding agents.
Context Engine — Connects AI tools and teams to one shared memory.
MS-Agent — It supports multimodal inputs, autonomous scheduling, and modular tool integration.
ShareCube — Publish agent artifacts to URLs for team annotation and editing.
Golem — Provides type-safe agent abstractions and zero-dependency Go implementation.
Claudian — Turns an Obsidian vault into a functional directory for AI agents.
LLM Wiki Compiler — Compiles raw sources into a citation-traced, interlinked markdown wiki.
Sparrow — Extracts structured data locally from documents without cloud dependencies.
MinerU — Parses complex documents and tables into structured markdown for agents.

All Trim 1 alternatives