Fix Continue + Ollama 'read_file' Invalid Arguments Error

Continue with a local Ollama model says read_file failed because the arguments were invalid (reading 'trim')? The model left out the file name. Here's the fix.

Luca Berton · The Production AI Expert

Published November 12, 2026·4 min read

You are in Continue’s Agent mode with a local Ollama model, you ask for a change, and instead of reading your file the agent stops with this:

read_file failed because the arguments were invalid, with the following message: Cannot read properties of undefined (reading 'trim')

Please try something else or request further instructions.

The short answer: this is not an Ollama bug, and nothing is wrong with your file. The model called Continue’s read_file tool without the file name. That is the classic weak spot of local models: tool calling is less reliable. In my setup it happens with gemma4:e4b, the small Gemma 4 model, which is fast but imprecise with tools.

The exact error messages this covers

read_file failed because the arguments were invalid, with the following message: Cannot read properties of undefined (reading 'trim')
The same message for other file tools, for example create_new_file failed because the arguments were invalid, with the following message: Cannot read properties of undefined (reading 'trim'), reported in Continue issue #12131.
A sibling from the same issue: edit_existing_file failed with the message: `filepath` and `changes` arguments are required to edit an existing file.

All of them end with “Please try something else or request further instructions.”

What the error actually means

Continue’s read_file tool has one required argument, filepath (definition). Before it runs a tool, Continue prepares its arguments. For read_file that means resolving the path, and the first line of the path resolver is inputPath.trim().

If the model’s tool call has no filepath (it sent empty arguments, or used another name such as path), the value is undefined, and JavaScript throws Cannot read properties of undefined (reading 'trim'). Continue catches it, marks the tool call as invalid instead of running it, and hands the message back to the model as the tool’s output: “tool failed because the arguments were invalid, with the following message: … Please try something else or request further instructions.”

So the error tells you about the model’s tool call, not about Ollama, the file or your workspace.

Fix it in this order

1. Use a model that is good at tools for Agent mode

In Continue’s model dropdown, at the bottom of the chat panel, switch Agent mode to a stronger tool caller. In my setup that means choosing “Qwen3.8 27B (Mac mini)”, not “Gemma4 fast”. Qwen is much more reliable with tools (it passed all the tool tests I ran with kagent), and it is slower.

Two checks for any model you try:

Does it support tool calling at all? ollama show <model> lists the model’s capabilities; you want tools among them. Ollama’s /api/show endpoint returns the same list as capabilities (API reference).
Does Continue know it? Continue detects tool support for most Ollama models by itself. With a custom model name, declare it in config.yaml, as Continue’s Ollama guide shows:

models:
  - name: Qwen3.8 27B (Mac mini)
    provider: ollama
    model: <your-qwen-model-tag>
    capabilities:
      - tool_use

Continue’s capabilities page adds that tool_use only switches on native tool calling: “The model must actually support tools for this to work.”

2. Retry the request

If Qwen also misses a tool call now and then, retry. Continue sends the error back to the model, and the model usually corrects itself on the next attempt.

3. The most reliable way: no tools at all

Use Chat (Cmd+L, or Ctrl+L on Windows and Linux) and attach the files yourself with @, for example @main.py, or select code and press Cmd+L. The model gets the content directly, without having to “read” it through a tool, and that works fine with Gemma too.

For changes, select the code and press Cmd+I (Ctrl+I): you describe the change, the model rewrites the selection, and no tool call is involved.

4. Give the model enough context

Continue sends a context size (num_ctx) with every Ollama request: the contextLength from your config.yaml, otherwise the model’s own num_ctx parameter, otherwise 8,192 tokens (source). Ollama’s own documentation recommends at least 64,000 tokens for agents and coding tools.

If your machine has the memory, raise it in Continue. Setting OLLAMA_CONTEXT_LENGTH on the server is not enough, because Continue sends its own value with each request:

models:
  - name: Qwen3.8 27B (Mac mini)
    provider: ollama
    model: <your-qwen-model-tag>
    capabilities:
      - tool_use
    defaultCompletionOptions:
      contextLength: 32768

Then check ollama ps: the CONTEXT column shows what the model got, and PROCESSOR shows whether it still fits on the GPU. If it spills to the CPU, go back down.

My split with local models

Chat and Cmd+I with @file work well and fast, Gemma included.
Agent mode only with Qwen, and expect the occasional retry.
For long, multi-file tasks, Claude stays clearly better.

FAQ

Is this an Ollama bug? No. Ollama returns what the model generated. The message comes from Continue checking the tool call before running it.

Why does it talk about trim? Because the first thing Continue does with the file path is inputPath.trim(). With no path in the tool call, there is nothing to trim.

Did it change my files? No. The tool call is marked as failed instead of running, and read_file only reads anyway.

Will a bigger context fix it with Gemma 4 E4B? It may help, but it doesn’t make a small model a reliable tool caller. For tool-heavy work, use the stronger model, or skip tools with @file.

Key takeaways

“read_file failed because the arguments were invalid … (reading ‘trim’)” means the model called read_file without filepath.
It is a tool-calling problem of the model. Small local models such as gemma4:e4b hit it most.
Fix it in this order: a stronger tool-calling model for Agent mode (Qwen in my case), a retry, or no tools at all: Chat and Cmd+I with @file.
Give Agent mode a bigger contextLength in Continue if your memory allows.

Sources

Continue source: preprocessToolCallArgs.ts, readFile.ts, pathResolver.ts, Ollama.ts
Continue docs: Ollama, model capabilities
Ollama docs: context length, show model details
Continue issue #12131: Qwen 3.5 and Gemma 4 can’t call tools