AXXA Agent by AXXA Lab™ Install Menu

obsidian plugin · desktop and mobile

Ask your vault. Tap through to the source.

One Obsidian panel, on your computer and your phone, for chat with the model you pick, answers that cite your own notes, and an agent that asks before it changes them.1

free · gpl-3.0 · your own key, or a local model · obsidian 1.11.4+

1 By default. You choose how much it does on its own, and while Obsidian is open, Undo reverts what it changed. How it asks first ↓

Pause motionPlay motion
AXXA Agent in Obsidian mobile. Asked what was decided with Northwind about the budget, Vault Q&A answers $6,400 and cites the note 2026-09-24 Northwind kickoff.
Fig. 1 Vault Q&A in Obsidian’s mobile layout. The answer cites a meeting note, and a tap opens it. Real recording, shortened: typing sped up, the model’s wait cut. Local model qwen3.5:9b on Ollama.

At a glance

01 answer and source

The answer, beside its source.

Ask in plain words. Vault Q&A answers from your notes and names the ones it used as links. Click one and the note opens; Ctrl/Cmd-click opens it in a new tab. On the phone, the panel steps aside so you can read it.

keyword
Names, dates, rare terms.
vector
Meaning, so other wording still matches.
links
A note linked from another match moves up.

Vector search uses an index stored in your vault, built with a cloud embedding model or one on your own Ollama. By default, re-indexing only re-embeds notes that changed. Without an index, search uses keywords.

Obsidian on the desktop. In the middle, the note 2026-09-24 Northwind kickoff lists the decision Budget approved at $6,400, paid in two halves. On the right, the AXXA Agent panel answers the budget question and cites that note.
Fig. 2 Obsidian on the desktop. In the middle, the meeting note: “Budget approved at $6,400.” On the right, the answer that cites it.

02 agent

It asks first. You can undo after.

Describe a change in plain words. The agent searches, reads, creates, edits, moves and deletes notes, and creates folders, all inside your vault. By default, every change waits for your OK and shows the tool and the exact path first.

ask
Askdefault

Every change waits for your OK.

vault
Vault

Creating, editing and moving run on their own. Deletes ask.

yolo
YOLO

Everything runs on its own, deletes included, to the trash your Deleted files setting picks. If Obsidian is set to delete files permanently, deletes still ask.

Review Agent change: the tool vault_move, from Personal/Lisbon trip.md to Travel/Lisbon trip.md, with the buttons Deny, Approve all and Approve.
Fig. 3 Before a move: the tool, the old path and the new one. Deny, Approve all (for the rest of this turn; deletes and web requests still ask) or Approve.
The agent's answer: the Lisbon trip note is now in the Travel folder. Under it, the count of actions it ran and an Undo button.
Fig. 4 After: Undo reverts the whole turn. The list of actions it ran has an Undo for each one.
  • Undo, per turn or per action.

    Undo under the answer reverts the whole turn, and each action has its own. It covers edits, new notes, moves, deletes and new folders while Obsidian stays open (a deleted file over 20 MB keeps no copy, so it only goes to your trash). If you edited the note by hand since, Undo stops and asks.

  • Obsidian does the moving.

    Moves go through Obsidian, so links update by your own link setting, and Undo writes them back. Deletes follow your Deleted files setting.

  • The web, on your terms.

    With Web access, on by default in Settings › AXXA Agent › Agent, the agent can open public pages, and search the web with your own Tavily key. In Ask and Vault, each search and address waits for your OK, unless you chose Approve all on a web request earlier in that turn.

03 in your notes

Right where you write.

Select text and right-click to rewrite it, fix its grammar and spelling, translate it, or send it to the chat. Right-click a note to ask about it or summarize it. Rewrites go straight into the note, and Ctrl/Cmd+Z takes them back.

7 commands
  • Ask about this note
  • Summarize this note
  • Send selection to chat
  • Rewrite selection
  • Fix grammar and spelling in selection
  • Translate selection…
  • Continue writing

On the phone, add them to the editor toolbar. In the chat, + › This note attaches the note you have open.

A line of the meeting note is selected, and the right-click menu lists Rewrite with AXXA, Fix grammar with AXXA, Translate with AXXA… and Send selection to AXXA.
Fig. 5 Right-click on selected text.
Add to Obsidian free, with your own key or a local model

04 questions

Four questions, answered.

Is it free?

Yes. No tier, no account, no license key. You pay your AI provider directly, or nothing on a free tier or a local model.

Will it change my notes without asking?

Not by default. The agent asks before every change, and Undo reverts what it did. How it asks ↑

Where do my notes go?

Only to the services you set up. Nothing goes to AXXA Lab™, and there is no telemetry. Privacy ↓

Does it work on my phone?

Yes. All three modes, confirmations included, run in Obsidian’s mobile app. Phone ↓

05 phone

Made for the phone first, not shrunk onto it.

All three modes, confirmations included, run in the Obsidian mobile app on Android and iOS. The desktop gets the same panel.

  • Above the keyboard.

    The composer stays above the on-screen keyboard.

  • Fullscreen, if you want it.

    An optional mode gives the panel the whole screen.

  • Same confirmations.

    The agent asks on the phone exactly as on the desktop.

AXXA Agent home in Obsidian's light theme: 57k tokens this month, a calendar of the month, recent Vault Q&A chats and buttons for Chat, Vault and Agent.
The same home in Obsidian's dark theme.
Figs. 6 and 7 Home, light and dark: this month’s tokens, a calendar of the month, recent chats and the three modes. Real captures, with Obsidian’s accent color set to graphite.

06 models and cost

Six providers. Your key. Your budget.

Use OpenAI, Anthropic, Google Gemini, OpenRouter or NVIDIA NIM with your own API key, or a model on your own Ollama server with no key. You pay the provider directly; AXXA Lab™ takes no cut.

  • OpenAI
  • Anthropic
  • Google Gemini
  • OpenRouter
  • NVIDIA NIM
  • Ollama local · no key
  • Free models, tagged.

    The model list tags each free model, with its limit where there is one, like free · 50/day on OpenRouter or free · 40/min on NVIDIA NIM.

  • Left today.

    What’s left on each free tier today, and when it resets.

  • A daily spending limit.

    A heads-up at 80% and 100%. Paid models can pause until midnight (Gemini counts at its paid price, even on the free tier); free and local models keep working.

  • Usage you can keep.

    Tokens or estimated cost by provider, model and mode, for all time or the last 7, 30 or 90 days. Report saves it as a Markdown note.

  • Five effort levels.

    Low to Max, each one tunable.

The Usage screen for October 2 to 6: total tokens and chats, split into sent, received and per chat, with a daily chart.
Fig. 8 Usage: tokens sent and received, by day, for all time or the last 7, 30 or 90 days.

Also: dictation (OpenAI) and read aloud (OpenAI, or your ElevenLabs voices), skills as slash commands, projects with their own instructions and source notes, and attachments: notes, images, a page by link, and PDFs with OpenAI, Anthropic or OpenRouter.

07 privacy

Your key. Your vault. Your machine.

  1. i

    Your key.

    Bring a key from OpenAI, Anthropic, Google Gemini, OpenRouter or NVIDIA NIM, or run Ollama with none. Keys are kept in Obsidian’s secret storage, not in the plugin’s settings file, so they don’t travel through Sync or backups.

  2. ii

    Your vault.

    Chats are saved as Markdown files in your vault, in a hidden folder by default. Each one also keeps a copy of what went with your messages (note excerpts, attached notes and pages, pasted text) and, in a long chat, the summary of its start. Skills, usage reports and exported chats are ordinary notes, and the search index lives in your vault too.

  3. iii

    Your machine.

    With Ollama, Chat, Vault Q&A and the agent can all run on your own computer. The agent needs a model that supports tools.

Your notes go only to the services you set up, never to AXXA Lab™. No telemetry, no AXXA account. Every request the plugin can make is listed in the Disclosures; this site’s own practices are in the privacy policy.

08 setup

Three steps to a first answer.

  1. Add it.

    Open the AXXA Agent page and choose Add to Obsidian, or search for AXXA Agent in Settings › Community plugins › Browse.

  2. Enable it.

    Install and enable it, then open Settings › AXXA Agent › Providers.

  3. Connect a provider.

    Add a key for one provider, or your Ollama address (usually http://localhost:11434). Then open the panel from the ribbon, or run AXXA Agent: Open panel.

Free ways to start

You only need one provider. Provider limits change; checked October 2026.

Google Gemini
A free tier on a project without billing. Image generation is paid.
OpenRouter
Free models, 50 requests a day per account (1,000 once the account has bought $10 in credits).
NVIDIA NIM
Free Endpoints on 20 hosted models, for development, 40 requests a minute.
Ollama
Runs models on your own computer, with no key and no account.

09 specification

What you’re installing.

modes
3Chat, Vault Q&A and Agent, in one panel.
in your notes
7 commandsEditor and right-click menus, mobile toolbar.
providers
6OpenAI, Anthropic, Google Gemini, OpenRouter, NVIDIA NIM, Ollama.
embeddings
5OpenAI, Google Gemini, NVIDIA NIM, OpenRouter, Ollama.
agent
8 + 2 toolsEight for the vault, two for the web. Ask (default), Vault or YOLO. Undo per turn or per action while Obsidian is open.
api keys
Your ownIn Obsidian’s secret storage, not in data.json. Ollama needs none.
storage
MarkdownChats as .md files in a hidden folder by default, with a copy of what went with each message. Skills, reports and exports are notes.
interface
EN · PT-BREnglish and Brazilian Portuguese, or the same as Obsidian.
price
FreeNo tier, no license key, no AXXA account.
license
GPL-3.0Source on GitHub.
requires
Obsidian 1.11.4+Windows, macOS, Linux, Android, iOS.
size
588 KBmain.js in 0.9.24 (587,963 bytes).
telemetry
NoneNothing is sent to AXXA Lab™.
review
PassedObsidian’s automated review. Health: Excellent.

588 KB of JavaScript.

The 12 most-downloaded AI plugins in the Obsidian directory have a median main.js of about 3.8 MB.

Fig. 9 Size of main.js. Measured October 2026.

Read the source on GitHub

11 faq

Common questions.

Is it free?

Yes. Every feature in the plugin today is free, with no tier, no account and no license key, and it stays that way. If paid options appear later, they will be new things built on top, never a lock on something that already worked. You pay your AI provider directly; the plugin takes no cut.

Which models can I use for free?

Each provider has its own rule. The model list in Settings tags each free model, with its limit where there is one, and the Usage screen shows what’s left today. Gemini: the Flash models, 2.5 Pro, embeddings and Gemma, on a project without billing, where Google may use what you send to improve its products; image generation is always paid. OpenRouter: models priced at zero, 20 requests a minute and 50 a day per account, or 1,000 a day once the account has bought $10 in credits. NVIDIA NIM: only the models NVIDIA marks Free Endpoint, for development and testing, up to 40 requests a minute. OpenAI: free daily tokens for eligible organizations that share their API data with OpenAI, on an account with a positive balance. Ollama: any local model. Checked October 2026.

Does it work on my phone?

Yes. Chat, Vault Q&A and the agent with its confirmations all run in Obsidian’s mobile app. Cloud providers work anywhere; Ollama runs on a computer, so a phone needs an Ollama server it can reach.

Can it run without the cloud?

Yes, with Ollama. Chat, Vault Q&A and the agent can all run on your own machine. The agent needs a model that supports tools.

What leaves my device?

Only what goes to the services you set up: your chat provider; your embedding provider, if it’s a different one; OpenAI for dictation; OpenAI or ElevenLabs for read aloud; Tavily and the pages the agent opens, when web access is used; and pages you attach by link. An image link inside an answer loads from wherever it points, and that server sees your IP address. Nothing goes to AXXA Lab™. The Disclosures list every request.

What if the agent gets it wrong?

Undo under the answer reverts the whole turn, and each action has its own, while Obsidian stays open. If you edited the note by hand since, Undo stops and asks first.

Is it in Portuguese?

Yes, the whole interface is available in Brazilian Portuguese. New installs start on Same as Obsidian, which follows Obsidian’s language. If you used AXXA before 0.9.23, choose Same as Obsidian or Português (Brasil) in Settings › AXXA Agent › Chat › Language.

Can I keep other AI plugins installed?

Yes. AXXA Agent runs in its own panel, its styles are scoped to that panel, and it doesn’t depend on or replace other plugins.

Where do I report a bug or ask a question?

Bugs go to the bug report form, questions and ideas to Discussions.

Try it on the vault you already have.

free · gpl-3.0 · your own key, or a local model