Anand Tech AI Assistant (Cloudflare AI)

A conversational AI assistant in the VS Code sidebar, using Cloudflare Workers AI free-tier models, for API and Azure API Management (APIM) help, code explanations, and Bicep / Terraform drafts for review.

From Anand Tech Inc — the team behind self-serve API onboarding for 120+ APIs, cutting implementation time from weeks to 20 minutes–1 day. Website: https://anandtechinc.com

This extension is a chat assistant, not a code-completion agent. It does not provide inline code completion like GitLab Duo or GitHub Copilot, and it does not edit your files automatically.

Full documentation is in DOCUMENTATION.md, included in the extension package. The changelog is in CHANGELOG.md, included in the extension package. The license is in LICENSE.txt, included in the extension package.

What it does

• Sidebar chat titled Anand Tech AI Assistant, with a model picker, message history for the current session only, a Send button, and a Clear chat button.
• Ask About Selection: send selected editor text with your question.
• Explain This Code: explain selected code in plain language.
• Generate APIM Onboarding Draft: answer prompts for API name, backend URL, product name, and Bicep or Terraform, and get a draft in a new untitled editor for review. The draft is never written to a file or applied by the extension. Review before use. Secrets must be placeholders. Deterministic generators in the other Anand Tech extensions remain the reliable path for production.

Setup — read this before first use

You need a Cloudflare account, an Account ID, and an API token. No card is required for the standard free allocation, according to Cloudflare's free-tier documentation checked 2026-10-27. See the sources section in DOCUMENTATION.md.

1. Create a free Cloudflare account at dash.cloudflare.com if you do not have one, and note your Account ID, shown on the Cloudflare dashboard.
2. Create an API token at dash.cloudflare.com, My Profile, API Tokens. Use the Workers AI template, which grants Workers AI Read permission. Copy the token once — you will not see it again.
3. In VS Code, run the command Anand Tech AI: Set Cloudflare Account ID, and enter your Account ID. The Account ID is stored in settings as a plain string. It is not a secret.
4. In VS Code, run the command Anand Tech AI: Set Cloudflare API Token, and enter your token. The token is stored ONLY in VS Code SecretStorage. It is never stored in settings.json, never written to a file, and never logged. Never paste the token in chat.
5. Open the sidebar view Anand Tech AI, pick a model, and send a message.

If setup is missing, the chat shows a friendly error telling you which command to run. It does not show a stack trace and it never shows your token.

Free-tier limits — please read

Cloudflare Workers AI free tier, as documented and checked 2026-10-27: 10,000 Neurons per day, shared across models, resetting daily at 00:00 UTC. Different models consume Neurons at different rates. Some newer frontier models explicitly require paid billing even though the free allocation exists, and model availability may change.

The model picker in this extension is labelled: Free-tier models (check Cloudflare billing — some models require paid billing). The default model is @cf/meta/llama-3.1-8b-instruct. The other models offered are @cf/meta/llama-3-8b-instruct, @cf/mistral/mistral-7b-instruct-v0.1, @cf/qwen/qwen1.5-7b-chat-awq, and @cf/openai/gpt-oss-120b.

If a model returns an error (404, 403, paid-billing required, or Neurons exhausted), the chat shows the exact error in plain language and suggests trying another model or checking the Cloudflare dashboard. The extension never silently falls back to a paid model.

This extension does not use the deterministic generators in the other Anand Tech extensions, and it does not replace them. For production APIM code, review any AI draft carefully, validate it with terraform plan or Bicep what-if, and prefer the deterministic APIM generator for repeatable output.

Privacy

Your messages and selected code are sent to Cloudflare Workers AI when you send a chat message. That is how the assistant works, and it is stated in the chat view itself: Free tier: 10,000 Neurons/day shared. Your messages and selected code are sent to Cloudflare Workers AI.

Chat history is kept in memory for the current session only and is NOT persisted to disk by default. Closing VS Code clears it. The extension makes no other network calls, has no telemetry, and has zero runtime dependencies. It uses only the built-in global fetch in the VS Code Node runtime.

Never send secrets, passwords, or tokens in chat. The assistant's system prompt tells it never to ask for them. If you accidentally paste a secret, rotate it immediately — chat history is in memory, but the message was already sent to Cloudflare.

Settings

Settings are under cloudflareAiAssistant in VS Code settings. There is NO token setting in configuration, by design.

• cloudflareAiAssistant.model — default @cf/meta/llama-3.1-8b-instruct
• cloudflareAiAssistant.accountId — your Cloudflare Account ID, plain string
• cloudflareAiAssistant.maxTokens — default 1024
• cloudflareAiAssistant.temperature — default 0.2, lower values give more focused answers for code and infrastructure drafts

Commands

• Anand Tech AI: Open Chat
• Anand Tech AI: Ask About Selection
• Anand Tech AI: Explain This Code
• Anand Tech AI: Set Cloudflare API Token
• Anand Tech AI: Set Cloudflare Account ID
• Anand Tech AI: Generate APIM Onboarding Draft

Troubleshooting

• 401 Authentication failed: the token was rejected. Check the token, its Workers AI Read permission, and expiry, then set it again with the Set Cloudflare API Token command.
• 403 Access denied: the model or account may require paid billing, or the token may lack permission. Try another model, check the Cloudflare dashboard and billing, and confirm the token uses the Workers AI template.
• 404 Model not found: try another model, for example the default Llama 3.1 8B model.
• 429 Rate limited or Neurons exhausted: the free daily allocation may be used up. Wait until 00:00 UTC, try a smaller request, or check usage on the Cloudflare dashboard.
• No reply at all: check your internet connection, confirm the Account ID and token are set, and check the Cloudflare dashboard status.
• Chat is empty after restarting VS Code: expected. History is in-session only and is not persisted.

Full troubleshooting, limitations, and the roadmap are in DOCUMENTATION.md, included in the extension package.

Limitations in 0.1.0

• No streaming. Replies arrive as one message after the model finishes.
• No live Cloudflare call was tested during development, because no token was available. Setup with your own token and Account ID is required to test it.
• No inline code completion, no automatic file edits, no multi-file agent workflows. This is a chat assistant.
• Model availability and Neuron costs may change. Check Cloudflare's current Workers AI documentation and billing pages.
• AI drafts may contain errors or invented details. Always review, validate with terraform plan or Bicep what-if, and never apply a draft containing real secrets.

License

MIT. See LICENSE.txt, included in the extension package. Copyright (c) 2026 Anand Tech Inc.