Search

import { nanoid } from "https://esm.sh/nanoid@5.0.5";
import { sqlite } from "https://esm.town/v/stevekrouse/sqlite";
import Anthropic from "npm:@anthropic-ai/sdk@0.24.3";
const TABLE_NAME = `memories`;
  try {
    // Get API key from environment
    const apiKey = Deno.env.get("ANTHROPIC_API_KEY");
    if (!apiKey) {
      console.error("Anthropic API key is not configured.");
      return null;
    }
    // Initialize Anthropic client
    const anthropic = new Anthropic({ apiKey });
    // Format previous facts for the prompt
    console.log({ message });
    const response = await anthropic.messages.create({
      model: "claude-3-5-sonnet-latest",
      max_tokens: 1000,

  project,
  branchId,
  // anthropicApiKey,
  // bearerToken,
  selectedFiles,

- [x] Add a "view source" / "send me a PR" link
- [x] Show the HTTP preview in second column if there is one (and let the user pick which one to preview in the iframe)
- [x] Figure out a convention to teach in the anthropic prompt mod where the LLM always checks the readme for the scope (if not provided) and keeps it up to take with every change
- [x] Ability to create new projects from the interface
- [x] Figure out why OpenTownie can't create HTTP vals. Maybe give it a seperate tool for it?
- [x] Start a timer for messages
- [x] Add more indicators that it's "still working"
- [x] Require users supply their own Anthropic token?
- [x] Add cost indications on messages
- [x] Add a bell noise when the message is done to let us know

import { readFile } from "https://esm.town/v/std/utils@71-main/index.ts";
import { createAnthropic } from "npm:@ai-sdk/anthropic@1.2.12";
import { ValTown } from "npm:@valtown/sdk@0.37.0";
import {
    project,
    branchId,
    anthropicApiKey,
    selectedFiles,
    images,
  // do we want to allow user-provided tokens still
  const apiKey = anthropicApiKey || Deno.env.get("ANTHROPIC_API_KEY");
  const our_api_token = apiKey === Deno.env.get("ANTHROPIC_API_KEY");
  if (our_api_token) {
    const traceId = `townie_${rowid}_${Date.now()}`;
    const anthropic = createAnthropic({ apiKey });
    // Wrap the Anthropic model with PostHog tracing
    tracedModel = withTracing(anthropic(model), phClient, {
      posthogDistinctId: distinctId,
      posthogTraceId: traceId,
    console.log("inside posthog" + tracedModel);
  } else {
    // Fallback to regular Anthropic call if PostHog is not configured
    const anthropic = createAnthropic({ apiKey });
    tracedModel = anthropic(model);
  }
    // @ts-ignore
    lastMessage.content.at(-1).providerOptions = {
      anthropic: { cacheControl: { type: "ephemeral" } },
    };
  }
          output_tokens: result.usage.completionTokens,
          cache_read_tokens:
            result.providerMetadata.anthropic.cacheReadInputTokens,
          cache_write_tokens:
            result.providerMetadata.anthropic.cacheCreationInputTokens,
        });
          output_tokens: result.usage.completionTokens,
          cache_read_tokens:
            result.providerMetadata.anthropic.cacheReadInputTokens,
          cache_write_tokens:
            result.providerMetadata.anthropic.cacheCreationInputTokens,
        });
      },

Townie is fully open-source and itself runs on Val Town. Pull requests welcome!
To get Townie running in your Val Town account, click the **Remix** button and then add your ANTHROPIC_API_KEY. You can leave all the other environment variables blank.
Authentication in Townie is handled via Val Town Oauth. However, we have not yet opened up our OAuth to anyone else, which currently makes it very awkward to use your own Townie. Here is a temporary workaround:

        </ul>
        <p>
          The application proxies requests to the Anthropic API and Val Town API, allowing Claude to view and edit your
          project files directly.
        </p>

    /** AI models, AI labs  */
    "https://openai.com/news/rss.xml",
    "https://raw.githubusercontent.com/Olshansk/rss-feeds/main/feeds/feed_anthropic_news.xml",
    "https://raw.githubusercontent.com/Olshansk/rss-feeds/main/feeds/feed_ollama.xml",
    "https://raw.githubusercontent.com/Olshansk/rss-feeds/main/feeds/feed_anthropic_news.xml",
    "https://raw.githubusercontent.com/Olshansk/rss-feeds/main/feeds/feed_anthropic_engineering.xml",
    "https://raw.githubusercontent.com/Olshansk/rss-feeds/main/feeds/feed_anthropic_research.xml",
    /** Individuals */

It is easy to reconfigure the agent to work with other providers that are supported by Vercel's AI SDK, though not all models support tool calling. 
Anthropic's models should work well, but the system prompt would need to be retuned to work with Claude.
Learn more: https://ai-sdk.dev/docs/foundations/providers-and-models

import { anthropic } from "npm:@ai-sdk/anthropic";
import { openai } from "npm:@ai-sdk/openai";
import { generateText, streamText } from "npm:ai";
  const maxSteps = 10;
  const model = Deno.env.get("ANTHROPIC_API_KEY") ? anthropic("claude-3-7-sonnet-latest") : openai("gpt-4.1");
  const options = {

It is easy to reconfigure the agent to work with other providers that are supported by Vercel's AI SDK, though not all models support tool calling. 
Anthropic's models should work well, but the system prompt would need to be retuned to work with Claude.
Learn more: https://ai-sdk.dev/docs/foundations/providers-and-models