https://www.boundaryml.com logo
🟡 Discussion on tool calling with BAML vs markdow...
# questions
m
Curious how folks think about tool calling with BAML and having the structured types be in the prompt vs something like how claude code has pure markdown. From the point of "if it's hard for humans to read, it's hard for LLMs to read/understand", then I feel like the specs should be more markdown than having 5-10 json
or
blocks, when you do something like BAML
type AgentAction = ReplyToUser | ReadFile | Glob...
Copy code
## Glob

**Description:**
Fast file pattern matching tool that works with any codebase size.
- Supports glob patterns like "**/*.js" or "src/**/*.ts"
- Returns matching file paths sorted by modification time
- Use this tool when you need to find files by name patterns
- When you are doing an open ended search that may require multiple rounds of globbing and grepping, use the Agent tool instead

**Parameters:**
- `pattern` (string, required): The glob pattern to match files against
- `path` (string, optional): The directory to search in. If not specified, the current working directory will be used.
v
claude code actually does both! it uses markdown + tool printing
codex does something similar
where they dump both into the prompt
i do agree that our union printing could be improved tho
esp for large unions
we have a
hoist_classes
option, so that helps some cases since you can just hosit some names up
m
ahh interesting. I'll look into hoisy classes. But thanks for the context!
quite interested in trying the or_splitter too
2 Views