beamcore/tools
The beamcore/tools model is a 0.3 billion parameter function-calling model developed by beamcore, fine-tuned from google/functiongemma-270m-it. It is specifically designed for on-device, local execution as a pre-flight search and routing assistant for the Elixir-native Beamcore agent harness. This lightweight model excels at analyzing user intent to determine if search/traversal tools are needed, emitting structured tool calls for local file system operations, and minimizing token usage for main reasoning models by providing relevant context.
Loading preview...
Overview
beamcore/tools is a highly optimized, lightweight (0.3 billion parameters) function-calling model, fine-tuned from google/functiongemma-270m-it. Its primary purpose is to act as a pre-flight search and routing assistant for the Elixir-native Beamcore agent harness, designed for efficient on-device execution.
Key Capabilities
- Local Execution: Optimized for running locally to reduce latency and cost.
- Pre-flight Search Routing: Analyzes user requests to determine if codebase context is needed before engaging a larger reasoning model.
- Structured Tool Calls: Emits specific tool calls for workspace inspection, including:
glob: Finds files matching a pattern.grep: Searches file contents by regex.tree: Shows a compact directory tree.read: Reads file contents with offset/limit.
- Token Minimization: By gathering necessary context locally, it drastically reduces token overhead and billing for subsequent calls to larger, more expensive frontier models.
Good For
- Agentic Software Engineering Workflows: Ideal for scenarios where large workspace structures and extensive tool schemas would otherwise lead to slow, expensive, and context-polluting interactions with large language models.
- Efficient Context Gathering: Automating the process of fetching relevant local files or code snippets based on user intent.
- Cost-Effective AI Agents: Reducing operational costs by minimizing the tokens sent to main reasoning models.