goose-agents

Personal Goose agent definitions — recipes, agents, and skills for the Goose AI agent framework.

Before installing: set your knowledge base path

Several files contain the placeholder /path/to/knowledge-base. You need to replace it with your actual knowledge base path before running install.

For example, if your knowledge base lives at /home/you/Nextcloud/Obsidian/Knowledge:

cd goose-agents
sed -i 's|/path/to/knowledge-base|/home/you/Nextcloud/Obsidian/Knowledge|g' \
  agents/*.yaml recipes/*.yaml skills/knowledge-base/SKILL.md

This replaces the placeholder in all agent configs and the knowledge-base skill. The agent-reach skill has no path references, so it doesn't need changing.

Installation

git clone https://github.com/Xarianne/goose-agents.git ~/GitHub/goose-agents
cd ~/GitHub/goose-agents

# Set your knowledge base path (see above)
sed -i 's|/path/to/knowledge-base|YOUR_KB_PATH|g' agents/*.yaml recipes/*.yaml skills/knowledge-base/SKILL.md

chmod +x install.sh
./install.sh

This copies files into ~/.agents/ and ~/.config/goose/skills/ (not symlinks). After editing repo files, re-run ./install.sh to update the installed copies.

Agents

Agent	What it does	When to use
Echo	Captures agent experiences, consolidates into compact judgment, reflects on past lessons	Session start (recall), session end (retain), weekly (consolidate)
Knowledge Curator	Writes/edits KB docs with proper frontmatter, citations, linking, log discipline	Any KB write or restructure
Search	Web research via SearXNG (privacy) + Tavily (AI-powered)	Any web search need

Configuration: provider and model

Each recipe YAML has a settings block with a provider and model. Change these to match your Goose setup. Edit the files in the repo, then re-run ./install.sh.

Via the Goose GUI

1. Open Goose and go to the Recipes tab
2. Click Edit on the recipe you want to change
3. Select your Provider and Model from the dropdowns
4. Save

Local models

If you run models locally (Ollama, LM Studio), select your local provider in the dropdown (e.g. ollama or lmstudio) and pick the model you have pulled. No API keys needed — the agents will call your local endpoint instead of a cloud API.

Manually (edit the YAML)

settings:
  goose_provider: ollama_cloud    # change this
  goose_model: gemma4:31b         # change this
  max_turns: 30

Current values (for reference)

Recipe	Provider	Model
echo	ollama_cloud	gemma4:31b
knowledge-curator	ollama_cloud	gemma4:31b
search	ollama_cloud	deepseek-v4-flash