chat-agent-toolkit 1.2.29 → 1.2.30

package/README.md

@@ -13,7 +13,7 @@ npm i chat-agent-toolkit
 ## Quick start
 
 ```ts
-import { writeLanguageResponse } from "ai-research-agent";
+import { writeLanguageResponse } from "chat-agent-toolkit";
 
 const response = await writeLanguageResponse({
   provider: "groq",
@@ -27,44 +27,27 @@ console.log(response.content);
 
 `response.content` is HTML by default (set `html: false` for raw Markdown). Agents that define an `after` hook also return parsed data on `response.extract`.
 
-## Supported providers
 
-| Provider        | ID           | Default model                                   | Notes                                             |
-| --------------- | ------------ | ----------------------------------------------- | ------------------------------------------------- |
-| Anthropic       | `anthropic`  | `claude-3-7-sonnet-20250219`                    |                                                   |
-| OpenAI          | `openai`     | `gpt-4o`                                        |                                                   |
-| Groq            | `groq`       | `meta-llama/llama-4-maverick-17b-128e-instruct` |                                                   |
-| Google Vertex   | `google`     | Gemini 2.x                                      |                                                   |
-| XAI             | `xai`        | `grok-beta`                                     |                                                   |
-| Amazon Bedrock  | `amazon`     | `anthropic.claude-3-5-sonnet-20241022-v2:0`     | apiKey = bearer token or `region:key:secret`     |
-| Cloudflare      | `cloudflare` | `llama-4-scout-17b-16e-instruct`                | apiKey = `token:accountId`                        |
-| Together AI     | `togetherai` | `meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo`   |                                                   |
-| Perplexity      | `perplexity` | `sonar`                                         |                                                   |
-| NVIDIA NIM      | `nvidia`     | `moonshotai/kimi-k2.5`                          |                                                   |
-| Ollama          | `ollama`     | `llama3.2`                                      | Local — no key required                           |
-
-The full registry (including context lengths) is exported as `LANGUAGE_MODELS`.
-
-### Amazon Bedrock setup
-
-Install the provider: `pnpm add ai @ai-sdk/amazon-bedrock`
-
-**Auth options** — set one of:
-- `AWS_BEARER_TOKEN_BEDROCK` (recommended API key auth)
-- `AWS_ACCESS_KEY_ID` + `AWS_SECRET_ACCESS_KEY` + `AWS_REGION` (SigV4)
-
-**Custom region:**
-
-```ts
-import { createAmazonBedrock } from '@ai-sdk/amazon-bedrock';
-
-export const amazonBedrock = createAmazonBedrock({
-  region: 'us-east-1',
-  apiKey: process.env.AWS_BEARER_TOKEN_BEDROCK,
-});
-```
+## Language Intelligence Providers
 
-> You must enable model access in the AWS Bedrock console before use; access is not granted by default.
+| Provider               | 🌍  | Top Model (Others)                            | 🏆 Benchmarks                                                         | 📄 Docs                                                                                               | 🔑 Keys                                                                                            | 💰 Funding   |
+| ---------------------- | --- | --------------------------------------------- | --------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------- | ------------ |
+| **Anthropic**          | 🇺🇸  | Claude Mythos / Opus  (Sonnet, Haiku)         | 🥇 GPQA Diamond 94.6% · 🥇 SWE-bench 93.9% · 🧬 PhD reasoning         | [Docs](https://docs.anthropic.com/en/docs/welcome)                                                    | [Keys](https://console.anthropic.com/settings/keys)                                                | ~$60B        |
+| **OpenAI**             | 🇺🇸  | GPT / o3 / Codex (o1, o4, o4-mini, gpt-4o)    | 🥇 AIME 2025 100% · 🥇 SWE-bench Pro · 📚 MMLU-Pro 90%                | [Docs](https://platform.openai.com/docs/overview)                                                     | [Keys](https://platform.openai.com/api-keys)                                                       | ~$180B       |
+| **Google**             | 🇺🇸  | Gemini Pro (Flash, Flash-Lite, Gemma)         | 🥇 GPQA 94.1% · 🥇 LiveCodeBench Elo 2439 · 🌐 #1 in 6/13 Vals        | [Docs](https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models)                            | [Keys](https://cloud.google.com/vertex-ai/generative-ai/docs/start/express-mode/overview#api-keys) | Public       |
+| **xAI**                | 🇺🇸  | Grok Heavy (Grok-3, Grok Vision)              | 🥇 AIME 2025 100% · 🧮 Math competition · ⚡ X integration            | [Docs](https://docs.x.ai/docs#models)                                                                 | [Keys](https://console.x.ai/)                                                                      | ~$45B        |
+| **Meta**               | 🇺🇸  | Llama Maverick / Scout (Llama 3.x, CodeLlama) | 🥇 DocVQA 94.4% · 🥇 10M token context · 📊 ChartQA 90%               | [Docs](https://www.llama.com/docs/overview/)                                                          | [Keys](https://www.llama.com/llama-downloads/)                                                     | Public       |
+| **NVIDIA**             | 🇺🇸  | Nemotron-Cascade  (Llama Nemotron, Kimi)      | 🥇 LCB v6 87.2% · 🏅 IMO+IOI+ICPC gold · 🧮 AIME 98.6%                | [Docs](https://docs.api.nvidia.com/nim/reference/llm-apis)                                            | [Keys](https://build.nvidia.com/settings/api-keys)                                                 | Public       |
+| **Perplexity**         | 🇺🇸  | Sonar Reasoning Pro (Sonar Deep Research)     | 🥇 Search Arena · 🔍 #1 web-grounded QA · 🌐 Real-time retrieval      | [Docs](https://docs.perplexity.ai/models/model-cards)                                                 | [Keys](https://www.perplexity.ai/account/api/keys)                                                 | ~$1B         |
+| **Groq**               | 🇺🇸  | (Llama, DeepSeek, Gemma, Mistral, Qwen)       | ⚡ #1 inference speed · 🏎️ Fastest TTFT · 🔧 LPU hardware             | [Docs](https://console.groq.com/docs/overview)                                                        | [Keys](https://console.groq.com/keys)                                                              | ~$640M       |
+| **Mistral**            | 🇫🇷  | Mistral Large  (Small 4, Codestral, Devstral) | 🥈 Arena Elo 1418 · 🌍 Multilingual MMLU 85.5% · 🚀 Fastest TTFT      | [Docs](https://docs.mistral.ai/)                                                                      | [Keys](https://console.mistral.ai/api-keys/)                                                       | ~$3.1B       |
+| **Together**           | 🇺🇸  | (Llama, Mistral, Gemma, Qwen, DeepSeek)       | 🏗️ Widest open hosting · 💸 Best open-source pricing · 🔧 Fine-tuning | [Docs](https://docs.together.ai/docs/quickstart)                                                      | [Keys](https://api.together.xyz/settings/api-keys)                                                 | ~$225M       |
+| **Moonshot**           | 🇨🇳  | Kimi Reasoning (K2.6, K2)                     | 🥇 AIME open 96.1% · 🥇 MATH-500 98% · 🥇 HumanEval 99%               | [Docs](https://platform.moonshot.cn/docs)                                                             | [Keys](https://platform.moonshot.cn/console/api-keys)                                              | ~$3.9B       |
+| **Zhipu**              | 🇨🇳  | GLM Reasoning / GLM-4.7 (GLM-4V, CogView)     | 🥇 Chatbot Arena Elo 1451 · 🥇 MMLU 96% · 🧮 AIME 95.7%               | [Docs](https://bigmodel.cn/dev/api)                                                                   | [Keys](https://bigmodel.cn/usercenter/apikeys)                                                     | ~$1.8B       |
+| **Alibaba**            | 🇨🇳  | Qwen-Coder / Qwen  (Qwen-VL, Qwen-Audio)      | 🥇 Codeforces Elo 2056 · 💻 SWE-bench 69.6% · 🏎️ LCB 70.7%            | [Docs](https://www.alibabacloud.com/help/en/model-studio/developer-reference/use-qwen-by-calling-api) | [Keys](https://bailian.console.aliyun.com/?apiKey=1)                                               | Public       |
+| **DeepSeek**           | 🇨🇳  | DeepSeek (DeepSeek-Coder, DeepSeek-VL)        | 🥇 IMO gold (open) · 📚 MMLU-Pro 81.2 · 🧮 AIME 87.5%                 | [Docs](https://api-docs.deepseek.com/)                                                                | [Keys](https://platform.deepseek.com/api_keys)                                                     | Bootstrapped |
+| **Cloudflare**         | 🇺🇸  | (Llama, Mistral, Gemma, Qwen, DeepSeek)       | 🌐 Edge inference · ⚡ Serverless CDN scale · 🔒 Privacy-first        | [Docs](https://developers.cloudflare.com/workers-ai/)                                                 | [Keys](https://dash.cloudflare.com/profile/api-tokens)                                             | Public       |
+| **Ollama**             | 🇺🇸  | (Llama, Mistral, Gemma, Qwen, DeepSeek)       | 🖥️ #1 local inference · 🔒 Fully offline · 🆓 Free self-hosted        | [Docs](https://ollama.com/docs)                                                                       | [Keys](https://ollama.com/settings/keys)   
 
 ## Built-in agents
 
@@ -100,7 +83,7 @@ import {
   SimpleMemory,
   MemoryAgent,
   DrizzleMemoryStorage,
-} from "ai-research-agent";
+} from "chat-agent-toolkit";
 ```
 
 - `SimpleMemory` — in-memory store implementing `IMemoryStorage`.
@@ -143,30 +126,10 @@ Vite bundles ES + CJS targets to `dist/`, emits `.d.ts` files alongside, and app
 - [Hugging Face tutorials](https://huggingface.co/learn)
 - [Illustrated Transformer](https://jalammar.github.io/illustrated-transformer/)
 - [Building a Transformer with PyTorch](https://www.datacamp.com/tutorial/building-a-transformer-with-py-torch)
-- [LLM training example](https://github.com/vtempest/ai-research-agent/blob/master/packages/neural-net/src/train/predict-next-word.js)
+- [LLM training example](https://github.com/vtempest/chat-agent-toolkit/blob/master/packages/neural-net/src/train/predict-next-word.js)
 
 <img src="https://i.imgur.com/uW6E9VJ.gif" alt="Transformer architecture visualization" />
-
-## Language Intelligence Providers
-
-| Provider               | 🌍  | Top Model (Others)                            | 🏆 Benchmarks                                                         | 📄 Docs                                                                                               | 🔑 Keys                                                                                            | 💰 Funding   |
-| ---------------------- | --- | --------------------------------------------- | --------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------- | ------------ |
-| **Anthropic**          | 🇺🇸  | Claude Mythos / Opus  (Sonnet, Haiku)         | 🥇 GPQA Diamond 94.6% · 🥇 SWE-bench 93.9% · 🧬 PhD reasoning         | [Docs](https://docs.anthropic.com/en/docs/welcome)                                                    | [Keys](https://console.anthropic.com/settings/keys)                                                | ~$60B        |
-| **OpenAI**             | 🇺🇸  | GPT / o3 / Codex (o1, o4, o4-mini, gpt-4o)    | 🥇 AIME 2025 100% · 🥇 SWE-bench Pro · 📚 MMLU-Pro 90%                | [Docs](https://platform.openai.com/docs/overview)                                                     | [Keys](https://platform.openai.com/api-keys)                                                       | ~$180B       |
-| **Google**             | 🇺🇸  | Gemini Pro (Flash, Flash-Lite, Gemma)         | 🥇 GPQA 94.1% · 🥇 LiveCodeBench Elo 2439 · 🌐 #1 in 6/13 Vals        | [Docs](https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models)                            | [Keys](https://cloud.google.com/vertex-ai/generative-ai/docs/start/express-mode/overview#api-keys) | Public       |
-| **xAI**                | 🇺🇸  | Grok Heavy (Grok-3, Grok Vision)              | 🥇 AIME 2025 100% · 🧮 Math competition · ⚡ X integration            | [Docs](https://docs.x.ai/docs#models)                                                                 | [Keys](https://console.x.ai/)                                                                      | ~$45B        |
-| **Meta**               | 🇺🇸  | Llama Maverick / Scout (Llama 3.x, CodeLlama) | 🥇 DocVQA 94.4% · 🥇 10M token context · 📊 ChartQA 90%               | [Docs](https://www.llama.com/docs/overview/)                                                          | [Keys](https://www.llama.com/llama-downloads/)                                                     | Public       |
-| **NVIDIA**             | 🇺🇸  | Nemotron-Cascade  (Llama Nemotron, Kimi)      | 🥇 LCB v6 87.2% · 🏅 IMO+IOI+ICPC gold · 🧮 AIME 98.6%                | [Docs](https://docs.api.nvidia.com/nim/reference/llm-apis)                                            | [Keys](https://build.nvidia.com/settings/api-keys)                                                 | Public       |
-| **Perplexity**         | 🇺🇸  | Sonar Reasoning Pro (Sonar Deep Research)     | 🥇 Search Arena · 🔍 #1 web-grounded QA · 🌐 Real-time retrieval      | [Docs](https://docs.perplexity.ai/models/model-cards)                                                 | [Keys](https://www.perplexity.ai/account/api/keys)                                                 | ~$1B         |
-| **Groq**               | 🇺🇸  | (Llama, DeepSeek, Gemma, Mistral, Qwen)       | ⚡ #1 inference speed · 🏎️ Fastest TTFT · 🔧 LPU hardware             | [Docs](https://console.groq.com/docs/overview)                                                        | [Keys](https://console.groq.com/keys)                                                              | ~$640M       |
-| **Mistral**            | 🇫🇷  | Mistral Large  (Small 4, Codestral, Devstral) | 🥈 Arena Elo 1418 · 🌍 Multilingual MMLU 85.5% · 🚀 Fastest TTFT      | [Docs](https://docs.mistral.ai/)                                                                      | [Keys](https://console.mistral.ai/api-keys/)                                                       | ~$3.1B       |
-| **Together**           | 🇺🇸  | (Llama, Mistral, Gemma, Qwen, DeepSeek)       | 🏗️ Widest open hosting · 💸 Best open-source pricing · 🔧 Fine-tuning | [Docs](https://docs.together.ai/docs/quickstart)                                                      | [Keys](https://api.together.xyz/settings/api-keys)                                                 | ~$225M       |
-| **Moonshot**           | 🇨🇳  | Kimi Reasoning (K2.6, K2)                     | 🥇 AIME open 96.1% · 🥇 MATH-500 98% · 🥇 HumanEval 99%               | [Docs](https://platform.moonshot.cn/docs)                                                             | [Keys](https://platform.moonshot.cn/console/api-keys)                                              | ~$3.9B       |
-| **Zhipu**              | 🇨🇳  | GLM Reasoning / GLM-4.7 (GLM-4V, CogView)     | 🥇 Chatbot Arena Elo 1451 · 🥇 MMLU 96% · 🧮 AIME 95.7%               | [Docs](https://bigmodel.cn/dev/api)                                                                   | [Keys](https://bigmodel.cn/usercenter/apikeys)                                                     | ~$1.8B       |
-| **Alibaba**            | 🇨🇳  | Qwen-Coder / Qwen  (Qwen-VL, Qwen-Audio)      | 🥇 Codeforces Elo 2056 · 💻 SWE-bench 69.6% · 🏎️ LCB 70.7%            | [Docs](https://www.alibabacloud.com/help/en/model-studio/developer-reference/use-qwen-by-calling-api) | [Keys](https://bailian.console.aliyun.com/?apiKey=1)                                               | Public       |
-| **DeepSeek**           | 🇨🇳  | DeepSeek (DeepSeek-Coder, DeepSeek-VL)        | 🥇 IMO gold (open) · 📚 MMLU-Pro 81.2 · 🧮 AIME 87.5%                 | [Docs](https://api-docs.deepseek.com/)                                                                | [Keys](https://platform.deepseek.com/api_keys)                                                     | Bootstrapped |
-| **Cloudflare**         | 🇺🇸  | (Llama, Mistral, Gemma, Qwen, DeepSeek)       | 🌐 Edge inference · ⚡ Serverless CDN scale · 🔒 Privacy-first        | [Docs](https://developers.cloudflare.com/workers-ai/)                                                 | [Keys](https://dash.cloudflare.com/profile/api-tokens)                                             | Public       |
-| **Ollama**             | 🇺🇸  | (Llama, Mistral, Gemma, Qwen, DeepSeek)       | 🖥️ #1 local inference · 🔒 Fully offline · 🆓 Free self-hosted        | [Docs](https://ollama.com/docs)                                                                       | [Keys](https://ollama.com/settings/keys)                                                           | ~$20M        |
+                                                        | ~$20M        |
 
 ## Alternative Agents Frameworks
 

package/dist/index.d.ts

@@ -13,6 +13,7 @@
  * @see {@link https://github.com/vtempest/ai-research-agent}
  */
 export * from 'write-language';
+export * from './prompts';
 export * from './memory';
 export * from './tools';
 export { AGENT_TOOLS } from './tools';

package/dist/prompts/index.d.ts

@@ -0,0 +1,5 @@
+/**
+ * @module agent-toolkit/prompts
+ * @description Prompt templates for AI agent tasks including web search and writing assistance.
+ */
+export * from './search-prompts';

package/dist/prompts/meta-search-types.d.ts

@@ -0,0 +1,6 @@
+/**
+ * @module agent-toolkit/prompts/meta-search-types
+ * @description Shared types for search prompt few-shot examples.
+ */
+/** A `[role, content]` tuple used for few-shot prompt examples. */
+export type FewShotExample = [role: "user" | "assistant", content: string];

package/dist/prompts/search-prompts.d.ts

@@ -0,0 +1,12 @@
+import { FewShotExample } from './meta-search-types';
+export declare const webSearchRetrieverPrompt = "\nYou are an AI question rephraser. You will be given a conversation and a follow-up question,  you will have to rephrase the follow up question so it is a standalone question and can be used by another LLM to search the web for information to answer it.\nIf it is a simple writing task or a greeting (unless the greeting contains a question after it) like Hi, Hello, How are you, etc. than a question then you need to return `not_needed` as the response (This is because the LLM won't need to search the web for finding information on this topic).\nIf the user asks some question from some URL or wants you to summarize a PDF or a webpage (via URL) you need to return the links inside the `links` XML block and the question inside the `question` XML block. If the user wants to you to summarize the webpage or the PDF you need to return `summarize` inside the `question` XML block in place of a question and the link to summarize in the `links` XML block.\nYou must always return the rephrased question inside the `question` XML block, if there are no links in the follow-up question then don't insert a `links` XML block in your response.\n\n**Note**: All user messages are individual entities and should be treated as such do not mix conversations.\n";
+export declare const webSearchRetrieverFewShots: FewShotExample[];
+export declare const webSearchResponsePrompt = "\n    You are QwkSearch, an AI model skilled in web search and crafting detailed, engaging, and well-structured answers. You excel at summarizing web pages and extracting relevant information to create professional, blog-style responses.\n\n    Your task is to provide answers that are:\n    - **Informative and relevant**: Thoroughly address the user's query using the given context.\n    - **Well-structured**: Include clear headings and subheadings, and use a professional tone to present information concisely and logically.\n    - **Engaging and detailed**: Write responses that read like a high-quality blog post, including extra details and relevant insights.\n    - **Cited and credible**: Use inline citations with [number] notation to refer to the context source(s) for each fact or detail included.\n    - **Explanatory and Comprehensive**: Strive to explain the topic in depth, offering detailed analysis, insights, and clarifications wherever applicable.\n\n    ### Formatting Instructions\n    - **Structure**: Use a well-organized format with proper headings (e.g., \"## Example heading 1\" or \"## Example heading 2\"). Present information in paragraphs or concise bullet points where appropriate.\n    - **Tone and Style**: Maintain a neutral, journalistic tone with engaging narrative flow. Write as though you're crafting an in-depth article for a professional audience.\n    - **Markdown Usage**: Format your response with Markdown for clarity. Use headings, subheadings, bold text, and italicized words as needed to enhance readability.\n    - **Length and Depth**: Provide comprehensive coverage of the topic. Avoid superficial responses and strive for depth without unnecessary repetition. Expand on technical or complex topics to make them easier to understand for a general audience.\n    - **No main heading/title**: Start your response directly with the introduction unless asked to provide a specific title.\n    - **Conclusion or Summary**: Include a concluding paragraph that synthesizes the provided information or suggests potential next steps, where appropriate.\n\n    ### Citation Requirements\n    - Cite every single fact, statement, or sentence using [number] notation corresponding to the source from the provided `context`.\n    - Integrate citations naturally at the end of sentences or clauses as appropriate. For example, \"The Eiffel Tower is one of the most visited landmarks in the world[1].\"\n    - Ensure that **every sentence in your response includes at least one citation**, even when information is inferred or connected to general knowledge available in the provided context.\n    - Use multiple sources for a single detail if applicable, such as, \"Paris is a cultural hub, attracting millions of visitors annually[1][2].\"\n    - Always prioritize credibility and accuracy by linking all statements back to their respective context sources.\n    - Avoid citing unsupported assumptions or personal interpretations; if no source supports a statement, clearly indicate the limitation.\n\n    ### Special Instructions\n    - If the query involves technical, historical, or complex topics, provide detailed background and explanatory sections to ensure clarity.\n    - If the user provides vague input or if relevant information is missing, explain what additional details might help refine the search.\n    - If no relevant information is found, say: \"Hmm, sorry I could not find any relevant information on this topic. Would you like me to search again or ask something else?\" Be transparent about limitations and suggest alternatives or ways to reframe the query.\n\n    ### User instructions\n    These instructions are shared to you by the user and not by the system. You will have to follow them but give them less priority than the above instructions. If the user has provided specific instructions or preferences, incorporate them into your response while adhering to the overall guidelines.\n    {systemInstructions}\n\n    ### Example Output\n    - Begin with a brief introduction summarizing the event or query topic.\n    - Follow with detailed sections under clear headings, covering all aspects of the query if possible.\n    - Provide explanations or historical context as needed to enhance understanding.\n    - End with a conclusion or overall perspective if relevant.\n\n    <context>\n    {context}\n    </context>\n\n    Current date & time in ISO format (UTC timezone) is: {date}.\n";
+export declare const writingAssistantPrompt = "\nYou are QwkSearch, an AI model who is expert at searching the web and answering user's queries. You are currently set on focus mode 'Writing Assistant', this means you will be helping the user write a response to a given query.\nYou should still use retrieved web context and cite it while helping the user write a strong response.\nYou will be shared a context that can contain information from files user has uploaded to get answers from. You will have to generate answers upon that.\n\nYou have to cite the answer using [number] notation. You must cite the sentences with their relevent context number. You must cite each and every part of the answer so the user can know where the information is coming from.\nPlace these citations at the end of that particular sentence. You can cite the same sentence multiple times if it is relevant to the user's query like [number1][number2].\nHowever you do not need to cite it using the same number. You can use different numbers to cite the same sentence multiple times. The number refers to the number of the search result (passed in the context) used to generate that part of the answer.\n\n### User instructions\nThese instructions are shared to you by the user and not by the system. You will have to follow them but give them less priority than the above instructions. If the user has provided specific instructions or preferences, incorporate them into your response while adhering to the overall guidelines.\n{systemInstructions}\n\n<context>\n{context}\n</context>\n";
+declare const _default: {
+    webSearchResponsePrompt: string;
+    webSearchRetrieverPrompt: string;
+    webSearchRetrieverFewShots: FewShotExample[];
+    writingAssistantPrompt: string;
+};
+export default _default;