Local Ollama with Web Search

Deutsche Fassung: README.de.md

An Obsidian plugin that embeds local AI models via Ollama directly into Obsidian. Chat assistant with web search, vault integration, persistent context and daily-note workflow. Fully local, no cloud, no API cost, no data sharing.

Features

Chat and web search

• Local queries – chat with your local model without any internet access (Ctrl+Enter or the Local Query button)
• Web search inside the chat – searches via DuckDuckGo, feeds the results back as context, answers with citations like [1], [2] (Shift+Enter or the Web Search button)
• Summarize note – condense the current note with one click
• Suggest structure – draft an outline for the active note
• Load note as context – attach background knowledge to the chat

Vault integration

• Save chat log – stores the full chat as a Markdown file. The model proposes folder, filename and tags; you confirm in a modal. Autocomplete over all existing vault folders, live preview, YAML frontmatter, AI callout, automatic filename suffix on conflicts.
• Daily Note – the model drafts a concise entry (H2 heading, at most two paragraphs) from the current chat and appends it to today's daily note. Path pattern with {YYYY}/{MM}/{DD} placeholders is configurable. Creates file and folder if needed.
• Load chat from note – re-open a saved Ollama chat file inside the chat view and continue the dialogue.

Prompt-driven save and folder creation

As of version 2.9.0 two filesystem actions can be triggered directly from the chat view through natural-language phrases, without opening the save modal.

Create a folder

Recognises phrasings such as:

• "create folder Archive/2026"
• "create a new folder 04 Resources/Drafts"
• "make a subfolder Raw data"
• "add folder 02 Projects/Book"

The path is interpreted relative to the vault root. Missing intermediate folders are created recursively. If the folder already exists the plugin reports that back as a notice and does not touch anything.

Save the chat or the last answer into a specific folder

Recognises phrasings such as:

• "save this to folder 04 Resources/Drafts"
• "save the last answer as Checklist to folder 01 Inbox"
• "put this in folder 02 Projects/Book"
• "store the result as Setup notes in folder 04 Resources/Drafts"

The target folder is created automatically if it does not exist yet. An optional filename can be passed with as …. Without a filename the plugin derives the title from the first few words of the first user prompt.

The word folder must appear in the prompt – without it the direct-save flow does not kick in and the save modal is shown instead.

Scope: full chat or last answer only

If the prompt contains a phrasing like "last answer", "last response", "the answer" or "the result", only the most recent question-and-answer pair is saved instead of the full history. In every other case the complete conversation is written.

Path validation

Every path is validated before any write. The following are rejected:

• empty paths
• absolute paths (leading /)
• relative segments (. or ..)
• segments containing < > : " | ? * or control characters

An invalid path produces an inline notice in the chat; nothing is written.

Conflict handling

If the target file already exists, the plugin shows an inline prompt in the chat with four options:

• Overwrite – replaces the entire file content with the new chat.
• Append – appends only the conversation part (from ## Conversation onwards) to the existing file, separated by a horizontal rule. The target file's frontmatter is left untouched.
• Keep both (rename) – writes a new file with suffix 2, 3, … in the same folder.
• Cancel – writes nothing.

The conflict prompt is not part of the saved chat and does not survive a restart.

Relation to the existing save flow

The Chat log header button and the generic phrases like "save that" still open the save modal with model suggestion, autocomplete and preview. The prompt-driven variants only kick in when a concrete folder is named in the prompt; otherwise the trigger logic falls back to the modal. All triggers share one toggle in Trigger phrases active.

German prompt phrases work equivalently. See README.de.md for the German examples.

Trigger phrases

Natural-language commands open the matching flow directly from the chat.

Toggle via the setting Trigger phrases active. German equivalents are documented in README.de.md.

Vault setup

• PARA folder structure – 00 Kontext, 01 Inbox, 02 Projekte, 03 Bereiche, 04 Ressourcen, 05 Daily Notes, 06 Archiv, 07 Anhänge. Alternative minimal template (just Inbox and Archive). Checkboxes per folder; existing folders are detected.
• Default context files – when 00 Kontext is created, the plugin writes three starter templates into it: Biographie.md, Schreibstil.md, Recherchevorgaben.md.
• Sort existing folders in – after the setup the plugin offers to move remaining top-level folders into the PARA structure or rename them. Assignment modal per external folder. Moves run through fileManager.renameFile, so wikilinks are updated automatically. Conflict dialog per case: merge, rename or skip.

Persistent context

Every .md file in 00 Kontext/ is automatically passed as background context to the model on every chat request. Filenames become subheadings, frontmatter is stripped. Hard cap at 8000 characters (configurable). This way the AI knows you, your writing style and your research preferences in every session.

Persistence

Messages and the loaded note context survive an Obsidian restart. The last chat is back as soon as you open the chat view (capped at the last 50 messages).

Requirements

1. Install Ollama

macOS / Linux:

curl -fsSL https://ollama.com/install.sh | sh

Windows: installer at ollama.com/download.

2. Pull a model

ollama pull gemma3:12b

Recommended models by hardware:

Note on model size: 7B and 8B models hallucinate more than 12B models. For reliable German output we recommend gemma3:12b on systems with at least 16 GB RAM.

3. Run Ollama

Ollama runs in the background after installation. Check with:

ollama list

Installation

Via the Obsidian Community Plugin store (recommended)

1. Obsidian → Settings → Community plugins → Browse
2. Search for "Local Ollama with Web Search"
3. Install → Enable

Updates will then be delivered automatically by Obsidian.

Via BRAT (for pre-store beta versions)

1. Install and enable the BRAT plugin in Obsidian
2. Open BRAT settings → Add Beta Plugin
3. Enter the URL https://github.com/Geolech/obsidian-local-ollama
4. Click Add Plugin → enable the plugin

Manual

1. Download the latest version from Releases
2. Copy main.js, manifest.json and styles.css into .obsidian/plugins/local-ollama-websearch/ inside your vault
3. Restart Obsidian → enable the plugin in Settings

Setup

1. Obsidian → Settings → Local Ollama
2. Pick a model from the dropdown (populated automatically from Ollama)
3. Vault setup (optional, one-off): the Create structure… button at the bottom of the settings tab creates the PARA folders and writes the starter files into 00 Kontext/
4. Fill in the context files: put your content into 00 Kontext/Biographie.md, Schreibstil.md, Recherchevorgaben.md – from that point on the AI knows you in every session

The plugin connects to http://localhost:11434 automatically – no further configuration needed.

Shortcuts and buttons

All actions are also available as command palette entries.

Settings (overview)

• Language – German or English
• Model – dropdown of all locally installed Ollama models
• System prompt – behaviour instruction, with a reset to default
• Save to vault – fallback folder, trigger phrases, default tags, open file after save
• Vault setup – PARA scaffolding and sorting existing folders
• Context folder – path, "load automatically" toggle, character cap, default files at setup
• Daily Note – path pattern with {YYYY} / {MM} / {DD} placeholders, append-by-default

Changelog

See CHANGELOG.md.

License

MIT – Frank Lechtenberg, TH OWL