🛠️ Tool: Read File 🛠️

The plainest of the three tool nodes in this pack: it gives an OllamaAgent the ability to read a local text file's contents and reason over them. Ask the agent a question about a document, a log, or notes you've got saved on disk, and - if it decides to call this tool - it opens the file and pulls the text back into the conversation instead of you having to paste it into the prompt yourself.

It's a narrow tool on purpose. It reads text files, full stop - no directory browsing, no searching across multiple files, no binary formats. If you need the agent to work from a document, this is what wires that up; if you need it to answer from live web data instead, that's OllamaTool_WebSearch's job.

How it works

The node's own description is specific about the one safety limit built in: content is capped at 10,000 characters to avoid overflowing the model's context window. So this isn't meant for feeding an agent an entire book - it's for a single, reasonably-sized document, note, or log file the agent needs to reference to answer a question.

The inputs and outputs that matter

There's exactly one input:

• tool_name (default read_local_file) - the name the agent sees internally when deciding whether to call this tool. The tooltip is the same guidance as the web search tool: rename it to something more descriptive for your use case (say, read_project_notes) if that helps a given model understand when to reach for it.

Notably, there's no file_path widget on this node itself - the path the tool reads presumably comes from how the agent invokes it during the conversation (the model decides what to pass based on your prompt), not from something you configure ahead of time on the node. That's consistent with how the other tool nodes work: they define a capability, and the model decides the specifics of how to use it at call time.

Output is a single tool value, type OLLAMA_TOOL - wire it into OllamaAgent's tools input directly, or into OllamaToolCombine if the agent needs this alongside other tools.

How to install it

Bundled with the rest of the pack. Via ComfyUI Manager, search "ComfyUI-Ollama-Describer" and install. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/alisson-anjos/ComfyUI-Ollama-Describer.git
pip install -r requirements.txt   # or install.bat on Windows

Restart ComfyUI. Needs Ollama running with a tool-calling-capable model (Llama 3.1 or Qwen 3.5, per the pack's own guidance) for the agent to reliably use this at all.

Common issues & troubleshooting

If the agent never actually reads the file, that's the same story as every other tool node here: it's most often a model that doesn't handle tool calling well, not a bug in this node - try Llama 3.1 or Qwen 3.5 before assuming something's broken. Given the 10,000-character cap, the other real trap is pointing this at a large file and getting a truncated, misleading read back - if you need an agent to work from something long, split it into smaller files or summarize it down first rather than relying on this tool to handle it whole. And because this node runs code that reads arbitrary paths off your filesystem based on what the model decides to pass, treat it the same way you'd treat any node that touches local files: don't hand an agent this tool in a workflow you're sharing or running against untrusted prompts without thinking about what it could be pointed at.