Ask questions and get answers from your Obsidian notes

Use Smart Chat API Extension when you want a written answer inside Obsidian based on notes you approve.

Smart Lookup can propose source notes.
You review the source list.
Smart Chat sends the approved sources and question to the configured model, then keeps the response in a saved Smart Chat thread.

What this guide covers

This guide is for Smart Chat API Extension, the API-model workflow supplied by Smart Chat Pro for configured local or cloud models inside Obsidian.

It is different from Smart Chat Core codeblocks.

For provider thread links in notes, use Getting Started with Smart Chat Core.

Before you start

You need:

• Smart Chat Pro installed and enabled
• Pro access for the API Extension workflow
• one configured chat model in Smart Environment
• provider credentials for a cloud model, or a running local model server for a local model
• Smart Environment ready enough for retrieval if you plan to use Lookup context

The first workflow does not require every model or source option to be configured.
One working model and one useful source are enough.

Confirm one model works

Open Smart Chat settings or Smart Environment settings.

Confirm that:

• a default chat model is selected
• the provider and model fields are complete
• the model test succeeds where a Test action is available
• a local model server is running when using a local provider

Do not continue configuring additional models before one model can answer a basic test request.

Get your first written answer from approved notes

1. Open Smart Chat API Extension

Click the Smart Chat ribbon icon shown below.

You should see a new thread or the current saved thread ready for input.

If the Smart Chat ribbon icon or view is absent:

1. Open Settings -> Community plugins.
2. Confirm Smart Chat Pro is installed and enabled.
3. Restart Obsidian after installing or updating it.
4. Search Smart Chat in the Command Palette for the open action supplied by your installed build.
5. If no Smart Chat API view or command exists, open Help from that build or report the installed Smart Chat Pro version.

Do not keep looking for an interface that is not present.

2. Confirm the model before sending

Use the model or status indicator to confirm which model will receive the request.

Current behavior can include:

• a default model selected in settings
• a model picker in the Smart Chat interface
• the provider and model recorded with each response

If the model is missing or unexpected, correct it before attaching sensitive sources.

3. Write one clear question

Start with one outcome-focused question.

Example:

What have I written about reducing information overload while researching?

Use only the notes attached to this request.
Name the notes or sections you relied on.
If the sources are insufficient, identify what is missing instead of guessing.

Other useful first questions:

Based on the attached sources, summarize the current state and list the next 5 actions.

Extract the constraints from these notes and propose a plan that satisfies every constraint.

Find contradictions across the attached sources and list what needs clarification.

Write the question before retrieving sources so the retrieval step has a clear target.

4. Choose how to attach sources

Smart Chat gives you two useful first-source paths.

Add context when you know the source

Use Add context when you already know which note should ground the answer.

This is the simplest first test because the source is known and easy to verify.

Good first sources include:

• the project brief that owns the question
• the decision note containing the relevant constraints
• the meeting note with the facts to summarize
• the draft you want reviewed

Use Lookup context when you know the question, not the notes

Use Lookup context when you know what you want to ask but do not know which notes contain the evidence.

The workflow is:

1. Write the question.
2. Run Lookup context.
3. Review the proposed sources.
4. Remove anything that should not guide the answer.
5. Send the request with the approved set.

Lookup proposes sources.
You decide what belongs in the request.

5. Review the source list before sending

Keep sources that contain:

• facts the answer needs
• prior decisions
• examples
• constraints
• current project state
• definitions that affect interpretation

Remove sources that are:

• only loosely related
• stale
• duplicated
• unreviewed or untrusted
• outside the scope of the question

Start with the smallest source set that can answer the question.
More context is not automatically better context.

6. Send the question

Send only after these are visible:

• the intended model
• the current question
• the approved source set

If no sources are attached, Smart Chat can offer a choice before sending.
Depending on the installed version, choices can include:

• Lookup context
• Select context
• Continue without context
• Cancel

Continuing without sources can be valid for general chat.
It is not a test of whether Smart Chat can answer from your notes.

7. Verify the response before trusting it

Compare the answer with the attached sources.

Check:

• Does the response use details that actually appear in the notes?
• Does it name the notes or sections it relied on when asked?
• Does it separate source-backed claims from suggestions?
• Does it identify missing information instead of filling gaps with confident guesses?
• Does the result satisfy the question you actually asked?

The response is working material, not automatically trusted note content.

You know it worked when

There are two proof layers.

A fluent response without visible source support is not the first win.

If the first answer does not work

The model is missing or fails

Use this recovery order:

1. Confirm the selected provider and model.
2. Retry the model test once.
3. Check cloud credentials or local server state.
4. Confirm the model is available to the configured account or local runtime.
5. Switch to another already-configured model if one is available.
6. Retry the same small request before changing the source set.

Do not change the model, prompt, and context at the same time.
Keep the test small enough to identify what failed.

Lookup context returns weak sources

1. Make the question more specific.
2. Add one distinguishing topic, project, person, or decision.
3. Remove instructions about the final report shape from the lookup wording.
4. Test the same query directly in Smart Lookup.
5. Confirm Smart Environment import and embedding coverage are ready enough.
6. Check whether relevant folders or files are excluded.

Lookup finds candidate sources.
It cannot recover a note that is excluded, not prepared, or described too vaguely.

The answer is generic

Before adding more notes:

1. Make the question or desired outcome explicit.
2. Add the most important constraint.
3. Remove weak or unrelated sources.
4. Confirm the attached source actually contains evidence for the question.
5. Ask the model to identify the exact missing source instead of guessing.

Recovery prompt:

Use only the attached sources.

First list the constraints and source notes you found.
Then answer the question.
If the sources are insufficient, identify the exact missing note, fact, or decision instead of guessing.

A generic answer often indicates an unclear request or weak source set, not a need for the entire vault.

The response contradicts a source

1. Open the source note.
2. Compare the exact passage with the response.
3. Ask Smart Chat to list the contradiction and quote only the relevant source section.
4. Correct the source set or question before asking for another full answer.
5. Keep the disputed response out of trusted note sections until reviewed.

The interface does not match this guide

Do not improvise from screenshots that describe another build.

Record:

• Smart Chat Pro version
• Smart Environment version
• Obsidian version
• desktop or mobile
• the missing button, action, or view

Then open Help from the installed plugin or report the mismatch.

What leaves your device

Local and cloud model configurations have different boundaries.

• A local provider can run on your machine when the selected local setup is running.
• A cloud provider receives the question and source content you choose to send through that provider-backed workflow.
• Smart Chat does not need to send your entire vault for each question.

Review the source list before sending sensitive material.
Use the smallest source set that can support the answer.

Core alternative: use the AI tool you already have

Smart Chat API Extension is not required to ask a model about selected notes.

Use this Core path when you prefer ChatGPT, Claude, Gemini, or another external AI tool:

1. Find likely sources with Smart Lookup.
2. Verify the useful notes.
3. Add them to Smart Context Builder.
4. Remove unrelated, stale, duplicated, or untrusted sources.
5. Copy the reviewed context bundle.
6. Paste it into your AI tool.
7. Ask the question and require the answer to identify its source notes.

Prompt starter:

Use only the attached notes as source material.

Question: [your question]

Name the notes or sections you relied on.
If the sources are insufficient, identify what is missing instead of guessing.

The same trust rule applies:
review the source set before sending and review the answer before promoting it into a trusted note.

After the first useful answer

Keep the next step tied to a real need.