Install Obsidian
Download and install Obsidian, then create or open the vault where you want to test ObservaStory. Obsidian is the writing surface; your scenes and story definitions remain Markdown files.
Learn
Install, learn, and evaluate stories in your own vault.
ObservaStory is not packaged as an Obsidian Community Plugin yet. These steps describe the manual path: install the pieces, connect a vault, run local evaluation, and confirm that results are written back into scene frontmatter.
Manual installation
Start here if you are new to this kind of setup.
The current workflow uses Obsidian for writing, community plugins for vault-side views and templates, Node scripts for evaluation, and Ollama for local model calls. Treat the first install as a small local toolchain, not a one-click plugin.
Before you begin: make a copy of your vault or start with a sample vault. The evaluator writes results into Markdown frontmatter, so test the workflow on disposable scenes first.
Enable the needed Obsidian plugins
Open Obsidian settings, allow community plugins, then install and enable Templater and Dataview. Templater runs vault commands and Dataview/DataviewJS powers many local views and reports.
Install Node.js
Install the current Node.js LTS release. Node runs the evaluation scripts, scheduler, claim collector, and utility commands outside Obsidian.
Install Ollama and pull a model
Install Ollama, start it, then pull the configured local model. The current example configuration uses qwen2.5:7b.
ollama pull qwen2.5:7b
Add the ObservaStory files
Copy the provided ObservaStory folders, templates, scripts, metric definition notes, and report files into your test vault or project folder. Keep the folder names from the package until you understand which paths are configurable.
Create your local configuration
Copy config.example.json to config.local.json. Set the Ollama URL, model name, vault paths, scheduler settings, and the list of evaluations you want to run.
Define the story objects
Create or copy notes for characters, arcs, plot threads, story engines, and any Truth Ledger claims. Scene notes should point to these objects in frontmatter so the evaluator knows which context belongs to the current scene.
Run one scene first
Before starting a batch, run evaluation on one disposable scene. Confirm that the script can reach Ollama and that the scene receives an ai: section in frontmatter.
Review the output
Check that scores look structurally correct before trusting them. Numeric results should land in predictable fields, and any rationale should match the configured rationale mode for that metric.
Move to batches and reports
After the single-scene test works, queue more scenes, start the scheduler, and use the reports to review the saved frontmatter values. Reports should read stored results instead of calling the model directly.
Vault setup walkthrough
Use this as the place for a guided walkthrough of folder structure, scene frontmatter, story object notes, and metric definitions.
Writing with feedback visible
Add the main demo video here with notes beneath it for context, version notes, and what the viewer should watch for.
Metric and report guide
Explain what each stored score means, where it comes from, and how a writer should interpret it.