OpenClaw + Ollama + Gemma setup on Windows

Hey everyone — I’m testing a local OpenClaw + Ollama + Gemma setup on Windows and wanted to see if anyone else has run into this behavior.

**Environment:**

* OS: Windows desktop

* Ollama: installed and working

* Model: `gemma4` / `gemma4:e4b`

* OpenClaw: installed and gateway reachable

* Local model responds correctly when running directly through Ollama

**What works:**

* `ollama --version` works

* `ollama list` shows installed models

* `ollama run gemma4:e4b` responds correctly

* `openclaw status --deep` shows the gateway reachable and event loop OK

* OpenClaw opens and the local agent can respond after disabling the web-search plugin issue

**Original issue:**

When using OpenClaw, I was getting this error:

```text

No callable tools remain after resolving explicit tool allowlist

tools.allow: *, ollama_web_search, ollama_web_fetch

no registered tools matched

Fix the allowlist or enable the plugin that registers the requested tool.

```

It looked connected to the `openclaw-web-search` plugin. My config had:

```json

"plugins": {

"allow": [

"openclaw-web-search",

"ollama",

"memory-core"

"entries": {

"ollama": {

"enabled": true

"openclaw-web-search": {

"enabled": true

}

"tools": {

"alsoAllow": [

"ollama_web_search",

"ollama_web_fetch"

"web": {

"fetch": {

"enabled": false

"search": {

"enabled": false

}

```

I temporarily fixed the tool error by disabling `openclaw-web-search` and removing `ollama_web_search` / `ollama_web_fetch` from the allowlist so I could run in local-only mode.

**Current issue:**

Now the local agent responds, but it has trouble consistently following a strict fixed output template. For example, I gave it a required review-report format with exact headings, but it keeps adding its own headings or summarizing freely instead of following the requested structure.

I’m trying to build a local-only, draft-only business document review workflow where the agent must:

* Review only the text provided

* Not invent context

* Not browse the web

* Not use tools

* Not add extra headings

* Follow one exact output format every time

**Question:**

Has anyone else seen this with OpenClaw + Ollama/Gemma, especially after disabling the web-search plugin? Is this mainly a model behavior issue, a session/context issue, or is there a better OpenClaw config pattern for strict local-only workflow agents?

Any recommended settings for:

* local-only agents

* no-tool mode

* strict output formatting

* resetting/clearing session context

* disabling broken plugin/tool references cleanly

would be appreciated.

0 comments