Skip to content
Change language: EN
My learningAdd to Chrome
BlogTutorials

How to use free AI inside AWFlow: Agents, Chat and Automation without an API key

Install a local AI model once and it powers everything in AWFlow: the chat assistant, your agents and your automation workflows. A practical walkthrough, step by step.

Dannick K.
October 14, 2026 · 2 min read
PresentBoard

Also in Français

A Pipeline block showing a web page going through Get All Text, then a local Web LLM model, and returning a 120-word summary in the AWFlow side panel.
Share
On this page
  1. Step 1 — Install a model
  2. Step 2 — Chat with free AI
  3. Step 3 — Build an agent that uses no API
  4. Step 4 — Add free AI to an automation workflow
  5. What free AI covers, and what it does not

Most AI tools charge per message. Every automation run, every chat reply, every agent action ticks a meter. AWFlow includes a local AI layer that runs models inside your browser. Install a model once, and the three main features—Chat, Agents and Automation—all use it at zero per-use cost.

This is a walkthrough. By the end you will have a model installed, the chat assistant pointing at it, an agent using it, and a working automation that summarises the page you are on.

Step 1 — Install a model

Everything starts on the Local AI page. Open AWFlow, go to Settings, then Local AI. You will see models grouped by task: chat, embeddings, image classification, speech-to-text and more.

The Local AI page: storage used by on-device models split by engine (WebLLM, TF.js, Transformers), a task list on the left (Chat, Embeddings, vision and audio tasks) and model cards on the right, such as Llama 3.2 1B Instruct with its size, variants and install state.The Local AI page: storage used by on-device models split by engine (WebLLM, TF.js, Transformers), a task list on the left (Chat, Embeddings, vision and audio tasks) and model cards on the right, such as Llama 3.2 1B Instruct with its size, variants and install state.

Pick a model and install it.For most laptops: Llama 3.2 1B (q4f16_1), about 879 MB.

Guided screenshot
Text version
  1. 1. Pick the task first: For chat and agents: pick Chat.
  2. 2. Read the card: Size and device fit before you download.
  3. 3. Protect your storage: Turn on Keep models installed.

For most laptops, start with Llama 3.2 1B (q4f16_1) — about 879 MB. If answers feel weak after testing, go to the 3B variant.

The detail view of Llama 3.2 1B Instruct on the Local AI page, scrolled to Use it in (Web LLM, Transformers Chat, Run Local Model) and Variants: four precisions (q4f16_1 ≈ 879 MB, q4f32_1 ≈ 1.1 GB, q0f16 ≈ 2.5 GB, q0f32 ≈ 5.0 GB), each with an Install button.The detail view of Llama 3.2 1B Instruct on the Local AI page, scrolled to Use it in (Web LLM, Transformers Chat, Run Local Model) and Variants: four precisions (q4f16_1 ≈ 879 MB, q4f32_1 ≈ 1.1 GB, q0f16 ≈ 2.5 GB, q0f32 ≈ 5.0 GB), each with an Install button.

Once done, AWFlow sets it as the default for chat.No further configuration needed.

Guided screenshot
Text version
  1. 1. Choose a variant: Smallest first: about 879 MB.
  2. 2. Install: Downloads in the background while you work.

Once the download finishes, AWFlow automatically sets the model as the default for chat.

Step 2 — Chat with free AI

Open the AWFlow side panel in any browser tab. The chat box is front and centre. The model picker at the bottom shows the model you just installed.

A new chat with Aria: the greeting, then the composer with the Ask Aria anything box, the Add context (+), Attach files and Dictate buttons, the Chat and Build switch, the context meter, the model picker set to Gemini Nano and the send button, above suggestion cards.A new chat with Aria: the greeting, then the composer with the Ask Aria anything box, the Add context (+), Attach files and Dictate buttons, the Chat and Build switch, the context meter, the model picker set to Gemini Nano and the send button, above suggestion cards.

Nothing left the tab.Zero data sent to an external endpoint when using Web LLM.

Guided screenshot
Text version
  1. 1. Check the model: Shows the installed model by default.
  2. 2. Type a message: Ask anything, or ask about the page.
  3. 3. Context meter: Shows how much of the window is used.

You can ask about the page you are reading. Type “Summarise this page” and AWFlow reads the visible text and passes it to the model.

Your browser
Local model
Chat panel
Web LLM modelrunning in your browser
page text + question
the summary
Network callnoneLeaves your deviceno

Same interface as a cloud model. Just a different destination.The chat box is identical; the difference is the model picker setting.

Round trip
Text version
  1. Your browser: Chat panel · Local model: Web LLM model
  2. Chat panel → Web LLM model: page text + question
  3. Web LLM model → Chat panel: the summary
  4. Network call: none · Leaves your device: no

Step 3 — Build an agent that uses no API

An agent is a chat-driven assistant that can act: read pages, fill forms, search notes. When the model behind the agent is local, every agent action is free.

The Agents tab of Agents & teams: a Describe an agent box with a Create agent button, the card for Aria (Your assistant, On-device AI) with a Chat button, and four templates to start from: Chief of staff, Inbox triager, Price watcher and Study buddy. Import and Blank agent buttons are at the top right.The Agents tab of Agents & teams: a Describe an agent box with a Create agent button, the card for Aria (Your assistant, On-device AI) with a Chat button, and four templates to start from: Chief of staff, Inbox triager, Price watcher and Study buddy. Import and Blank agent buttons are at the top right.

Either way, the model is the one you installed.The agent uses the Local AI default unless you change the model picker.

Guided screenshot
Text version
  1. 1. Start blank: Or pick a template.
  2. 2. Describe what you want: Aria drafts the agent for you.
  3. 3. Use a template: Pre-built agents for common jobs.

In the agent builder, the Identity tab sets the role and name. The Skills tab controls what the agent is allowed to do. After saving, the agent uses the default local model. To override it for one agent, change the model picker inside that agent’s chat:

The Assistant page with a saved chat open: the user asks what a webhook is and whether AWFlow can start a workflow from one, and Aria answers in two short replies. The left rail lists recent chats.The Assistant page with a saved chat open: the user asks what a webhook is and whether AWFlow can start a workflow from one, and Aria answers in two short replies. The left rail lists recent chats.

Switch back to the default at any time.The picker resets to the Local AI default if you clear the override.

Guided screenshot
Text version
  1. 1. Open an agent: Each agent has its own thread.
  2. 2. Model for this agent: Change it here to override the default.
  3. 3. Chat: All its actions use the chosen model.

Step 4 — Add free AI to an automation workflow

Automations run on a schedule or on a trigger, without you being there. A local model works in automations too: add a Web LLM node, and every run is free.

The fastest way to see this is the Summarize This Page example. It reads the page you are on and returns a summary in the side panel.

Summarize This Page4 steps · AWFlow docs
Copy to my AWFlow (opens in a new tab)
TriggerRun now
PageRead the current page
AISummarize it
PageShow the summary
Summarize itMax words120Stylea few bullet points

Open the summary step. Change length and style to taste.Summarize node: max 120 words, a few bullet points. The model is the Local AI default.

Workflow embed
Text version
  1. summarize-this-page.awf
  2. Max words: 120 · Style: a few bullet points

After copying, look at the Summarize step. Its model slot is pre-filled with the model you installed.

The workflow editor with the Summarise this page in 3 bullets example: a sticky note, then When Started, Get All Text, Basic LLM Chain and Display Markdown in a row. A Web LLM model node hangs under the chain’s Model slot. The chain has a warning badge and the Ready to run? button counts one setting still to fill.The workflow editor with the Summarise this page in 3 bullets example: a sticky note, then When Started, Get All Text, Basic LLM Chain and Display Markdown in a row. A Web LLM model node hangs under the chain’s Model slot. The chain has a warning badge and the Ready to run? button counts one setting still to fill.

The model slot is the only thing that changes if you upgrade.The logic, the prompt and the rest of the workflow stay exactly the same.

Guided screenshot
Text version
  1. 1. The AI step: Holds the prompt.
  2. 2. Model slot: Your installed local model lives here.
  3. 3. Test run: Runs on the current page.

The model slot is intentional. The workflow does not hardcode a model: it uses whichever model is in the slot. Swap to a cloud model for tougher jobs; the rest of the workflow stays the same.

Any web page
Get All Text
Summarize (Web LLM)
Show Summary
side panel result

The result appears in the side panel.Show Summary node writes to the display panel.

Pipeline
Text version
  1. Any web page: visible text
  2. Get All Text: raw page text
  3. Summarize (Web LLM): 120-word summary
  4. Show Summary: side panel result

What free AI covers, and what it does not

Local models handle a wide range of everyday AI jobs well. They miss on tasks that need large context windows, the most recent knowledge or very complex reasoning.

Good fits for local models

What a local model handles well in AWFlow

  • Summarising pages and articles (done)local AI
  • Extracting structured fields from text (done)local AI
  • Classifying and tagging items (done)local AI
  • Rewriting and reformatting text (done)local AI
  • Drafting short messages based on context (done)local AI
  • Answering questions about the current page (done)local AI

Q&A on the current page is the simplest workflow you can build.Pass the page text as context; ask the question.

Checklist
Text version
  1. What a local model handles well in AWFlow
  2. Summarising pages and articles
  3. Extracting structured fields from text
  4. Classifying and tagging items
  5. Rewriting and reformatting text
  6. Drafting short messages based on context
  7. Answering questions about the current page

When a task regularly produces weak results, the upgrade path is one change: swap the model slot to a cloud provider. The rest of the workflow is unchanged, and only that step becomes a paid call.

Your turn

Try it: Summarize any page for free

Copy the workflow, install a local model, run it on the page you are reading.

Open in AWFlow (opens in a new tab)docs.awflow.io/examples/summarize-this-page.awf
Scan to open

Try it now. The workflow is one click away.

Try it
Text version
  1. Try it: Summarize any page for free · Copy the workflow, install a local model, run it on the page you are reading. · https://docs.awflow.io/examples/summarize-this-page.awf

What we covered

  1. Install once on Local AI: chat, agents and workflows share the model
  2. The chat assistant uses the local model by default
  3. Workflows use a model slot: free now, cloud model later, no rewrite

Workflows stay the same when you upgrade the model.The model slot is a dependency node; changing it does not change the logic.

Recap
Text version
  1. What we covered
  2. Install once on Local AI: chat, agents and workflows share the model
  3. The chat assistant uses the local model by default
  4. Workflows use a model slot: free now, cloud model later, no rewrite
Written by
Dannick K.
All articles

Keep reading

Back to the blog →
Tutorials

Use Your Web Browser to Feed a RAG Pipeline

Build a RAG knowledge base from web pages without a crawler or a vector server: save pages and tabs from your browser, index them on your device, then ask questions.

5 min readDecember 8, 2026
Tutorials

Browser Automation Without Selenium, Playwright or Puppeteer

Map each Selenium, Playwright or Puppeteer step (open, click, type, wait, extract, schedule) to a no-driver workflow in your own Chrome, and when to keep the script.

6 min readDecember 1, 2026
Tutorials

No-Code Browser Automation: Record Once, Replay Forever

Record your clicks and typing once in Chrome, then replay them for every row of a sheet, from a shortcut or on a schedule. No code, no server.

8 min readNovember 17, 2026

Automate the web you're on. Free, in your browser.

Add to Chrome — free