From 576923ddaa304e27b4e1cdef5195021b85268cb5 Mon Sep 17 00:00:00 2001 From: Thibault Molleman Date: Tue, 10 Dec 2024 21:53:41 +0100 Subject: [PATCH 1/7] Changes to assistant --- docs/kagi/ai/assistant.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/kagi/ai/assistant.md b/docs/kagi/ai/assistant.md index 91a6841d..e35503ae 100644 --- a/docs/kagi/ai/assistant.md +++ b/docs/kagi/ai/assistant.md @@ -28,7 +28,7 @@ The Assistant by Kagi combines top large language models (LLM) with optional res ## Privacy -When you use the Assistant by Kagi, your data is never used to train AI models, and no account information is shared with the LLM providers. By default, threads expire after 24 hours of inactivity. This behavior can be adjusted in the [settings](https://kagi.com/settings/?p=assistant). +When you use the Assistant by Kagi, your data is never used to train AI models (not by us or by the LLM providers), and no account information is shared with the LLM providers. By default, threads expire after 24 hours of inactivity. This behavior can be adjusted in the [settings](https://kagi.com/settings/?p=assistant). ## Using the Assistant From a34969ee5e19ad0bb2fe3cbe220ed802849e0d4e Mon Sep 17 00:00:00 2001 From: Thibault Molleman Date: Tue, 10 Dec 2024 22:07:42 +0100 Subject: [PATCH 2/7] Rewrite of Bangs section to make it clear that Chat is without internet --- docs/kagi/ai/assistant.md | 13 ++++++++++--- 1 file changed, 10 insertions(+), 3 deletions(-) diff --git a/docs/kagi/ai/assistant.md b/docs/kagi/ai/assistant.md index e35503ae..d40a6f72 100644 --- a/docs/kagi/ai/assistant.md +++ b/docs/kagi/ai/assistant.md @@ -160,10 +160,17 @@ You can learn more about how these models compare in the [Kagi LLM Benchmarking ## Bangs -You can quickly access Assistant from Kagi Search by using the following [bangs](../features/bangs.md): -`!ai`, `!chat`, `!assistant` , `!llm`, `!asst`, `!as`, `!expert` and `!fast` +You can quickly access the Kagi Assistant from Kagi Search using the following [bangs](../features/bangs.md): -You can also access the **Code** Custom Assistant with the `!code` bang. +- `!ai`, `!as`, `!assistant`, `!research`, `!answer`, `!discuss`, `!expert`, `!llm`, `!custom`, and `!asst`: These bangs direct you to the general Kagi Assistant interface for various types of queries. + +- `!chat`: This bang accesses the Kagi Assistant with internet access turned off. + +- `!code`: Use this bang to access the **Code** Custom Assistant, which is tailored for coding-related queries. + +- `!ki`: This bang accesses the Kagi Assistant with the Ki profile, providing a specialized interaction. + +Each bang is designed to optimize your search experience by directing you to the most appropriate version of the Kagi Assistant for your needs. ## URL Parameters From 6888e309c83d54cc54cec92ccb736dd051f1e17d Mon Sep 17 00:00:00 2001 From: Thibault Molleman Date: Tue, 10 Dec 2024 22:31:05 +0100 Subject: [PATCH 3/7] Clarified thread downloading --- docs/kagi/ai/assistant.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/kagi/ai/assistant.md b/docs/kagi/ai/assistant.md index d40a6f72..ff40b72b 100644 --- a/docs/kagi/ai/assistant.md +++ b/docs/kagi/ai/assistant.md @@ -55,7 +55,7 @@ By default threads are kept for 24 hours after the last message. If keeping threads alive permanently better fits your workflow, you can adjust this setting in [Assistant Settings](https://kagi.com/settings/?p=assistant). Please note that the thread saving setting is applied **when the thread is created**. -Threads can be renamed, added to Quick Access, shared, deleted, and downloaded via the `⋮` button which is displayed when you hover over the thread. +Threads can be renamed, added to Quick Access, shared, deleted, and downloaded via the `⋮` button which is displayed when you hover over the thread. When you download a thread, it is saved as a `.txt` file containing only the most recent path, which is the sequence of messages from your latest interaction in that thread. This path reflects the latest direction the conversation took based on your inputs. The Quick Access feature can be used to better organize your threads but it will not influence the thread saving setting. In other words, threads added to Quick Access will also be deleted after 24 hours, unless you have chosen otherwise in the settings. From c7c01a23c9464872da2aca1e7f10d448db6f6c6b Mon Sep 17 00:00:00 2001 From: Thibault Molleman Date: Tue, 10 Dec 2024 23:22:31 +0100 Subject: [PATCH 4/7] link to custom assitants page and clarify code being built-in --- docs/kagi/ai/assistant.md | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/docs/kagi/ai/assistant.md b/docs/kagi/ai/assistant.md index ff40b72b..25b8bf64 100644 --- a/docs/kagi/ai/assistant.md +++ b/docs/kagi/ai/assistant.md @@ -111,7 +111,9 @@ You can, for instance, instruct the Assistant to be more succinct or to consider You can create Custom Assistants in the [Assistant Settings](https://kagi.com/settings?p=assistant). It is possible to customize the LLM, settings (the use of web access, lenses, and personalized results), and the instructions for each Custom Assistant. -The Assistant comes with a **Code** Custom Assistant that is optimized for programming tasks. It uses Claude 3.5 Sonnet and has web access. +The Assistant comes with a built-in **Code** Custom Assistant that is optimized for programming tasks. It uses Claude 3.5 Sonnet and has web access. + +For more details, refer to the [Custom Assistants](./custom-assistants.md) page. ## Keyboard Shortcuts {#keyboard-shortcuts} @@ -166,7 +168,7 @@ You can quickly access the Kagi Assistant from Kagi Search using the following [ - `!chat`: This bang accesses the Kagi Assistant with internet access turned off. -- `!code`: Use this bang to access the **Code** Custom Assistant, which is tailored for coding-related queries. +- `!code`: Use this bang to access the built-in **Code** Custom Assistant, which is tailored for coding-related queries. - `!ki`: This bang accesses the Kagi Assistant with the Ki profile, providing a specialized interaction. From 50c5b2eeae231f842d6bb1d2040a2f53cbdd939e Mon Sep 17 00:00:00 2001 From: Thibault Molleman Date: Wed, 11 Dec 2024 09:06:10 +0100 Subject: [PATCH 5/7] added Qwen, updated llm list, add info about model hosting --- docs/kagi/ai/assistant.md | 15 +++++++++++---- 1 file changed, 11 insertions(+), 4 deletions(-) diff --git a/docs/kagi/ai/assistant.md b/docs/kagi/ai/assistant.md index 25b8bf64..d9acfe98 100644 --- a/docs/kagi/ai/assistant.md +++ b/docs/kagi/ai/assistant.md @@ -144,19 +144,26 @@ The following keyboard shortcuts are available in the Assistant by Kagi on Mac a The Assistant provides access to the following LLMs: - Anthropic - - Claude 3 Haiku - Claude 3.5 Sonnet - - Claude 3 Opus + -Claude 3 Opus + - Claude 3 Haiku - OpenAI - GPT 4o - - GPT 4o Mini + - GPT 4o mini - Mistral AI - Mistral Pixtral - Mistral Large - Google - - Gemini 1.5 Pro + - Gemini Pro - Meta + - Llama 3.3 70B - Llama 3.1 405B +- Qwen + - Qwen QwQ 32b + +For our platform, Llama and Qwen models are hosted by Fireworks.ai. The other models are accessed through APIs provided by their respective companies. + +In either case, none of these the data you submit is contractually allowed to be used for training purposes. You can learn more about how these models compare in the [Kagi LLM Benchmarking Project](./llm-benchmark.md) page. From 805914ebd5d1dc6ccd132a3f3cb676f999d8ef25 Mon Sep 17 00:00:00 2001 From: Thibault Molleman Date: Wed, 11 Dec 2024 09:10:51 +0100 Subject: [PATCH 6/7] fix to previous commit --- docs/kagi/ai/assistant.md | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/docs/kagi/ai/assistant.md b/docs/kagi/ai/assistant.md index d9acfe98..ab6a231f 100644 --- a/docs/kagi/ai/assistant.md +++ b/docs/kagi/ai/assistant.md @@ -145,16 +145,16 @@ The Assistant provides access to the following LLMs: - Anthropic - Claude 3.5 Sonnet - -Claude 3 Opus + - Claude 3 Opus - Claude 3 Haiku - OpenAI - GPT 4o - - GPT 4o mini + - GPT 4o Mini - Mistral AI - Mistral Pixtral - Mistral Large - Google - - Gemini Pro + - Gemini 1.5 Pro - Meta - Llama 3.3 70B - Llama 3.1 405B From 856c5f8c2139764fe1858a34b91d7faed8743b18 Mon Sep 17 00:00:00 2001 From: Thibault Molleman Date: Wed, 11 Dec 2024 09:35:40 +0100 Subject: [PATCH 7/7] v1 is sunset --- docs/kagi/ai/assistant_v1.md | 97 ------------------------------------ 1 file changed, 97 deletions(-) delete mode 100644 docs/kagi/ai/assistant_v1.md diff --git a/docs/kagi/ai/assistant_v1.md b/docs/kagi/ai/assistant_v1.md deleted file mode 100644 index 70ce4219..00000000 --- a/docs/kagi/ai/assistant_v1.md +++ /dev/null @@ -1,97 +0,0 @@ -# Assistant - -Kagi Assistant is a product feature backed by Kagi Search and large language models. - -**Kagi Assistant is currently in open beta and available to [Ultimate plan](https://kagi.com/pricing) members.** - -## Assistant Modes - -### Research - -Research assistant uses Kagi search and external sources like Wolfram Alpha, together with an AI model to answer your queries. -This model also allows you to upload a file (up to 16 MB) or provide a URL (article, PDF, video, podcast, etc.) to use as context. -If an image is provided it will be processed using GPT4 Vision. - -![Kagi Assistant - Research](./media/assistant_research.png){width=500px data-zoomable} - -This mode is available in two variants: - -- `Fast`: Kagi search coupled with a fast LLM (`Claude 3 Haiku`). Optimized for speed, while not sacrificing quality. Access directly with the `!fast` bang. -- `Expert`: Multiple Kagi searches (and external services) together with a better LLM (`GPT-4o`) for the highest accuracy and comprehensive responses. Access directly with the `!expert` bang. - -Research Assistant supports using [Lenses](../features/lenses.md) to filter the types of websites used as context to generate an answer. Lenses are predefined categories that help you focus on specific topics or sources. This allows you to restrict results to more trustworthy or relevant sites for your needs. To enable this feature, toggle the Lenses switch and select one of the available options. - -![Research Assistant - Lenses](./media/assistant_research_lenses.png){width=500px data-zoomable} - - -### Code - -Code assistant is tuned for programming tasks and uses a powerful AI model for programming. - -![Kagi Assistant - Code](./media/assistant_code.png){width=500px data-zoomable} - -This mode uses `gpt-4o`. - -Access directly with the `!code` bang. - -### Chat - -Chat assistant allows direct access to a variety of large language models for creative tasks. - -![Kagi Assistant - Chat](./media/assistant_chat.png){width=500px data-zoomable} - -This mode supports the following models: - -- OpenAI - - GPT 3.5 Turbo - - GPT 4 - - GPT 4 Turbo - - GPT 4o -- Anthropic - - Claude 3 Haiku - - Claude 3.5 Sonnet - - Claude 3 Opus -- Google - - Gemini Pro -- Mistral AI - - Mistral Small - - Mistral Large - -Access directly with the `!chat` bang. - -![LLMs supported by Kagi Assistant](https://kagifeedback.org/assets/files/2024-03-26/1711457673-662718-image.png) - -### Custom - -Custom assistant allows you to further customize one of the existing modes. - -![Kagi Assistant - Custom](./media/assistant_custom.png){width=500px data-zoomable} - -Select the base mode (and variant/model where applicable) and add your own custom instructions. Setup is detailed in the [Assistant settings help page](../settings/assistant.md#custom-assistant). - -Access directly with the `!custom` bang. - -## Tips and Tricks for using Kagi Assistant - -- Assistant has a character limit of 16,000 for each prompt in a conversation. -- You can input math equations using the LaTeX format. -- Using the copy button will copy the response to your clipboard preserving any formatting using markdown. -- Assistant supports syntax highlighting in generated code snippets. -- For optimal results, interact with the Assistant in a natural language. For instance, "Who was Alan Turing?" will yield a better response than simply "Alan Turing". -- When using Code Assistant, it is recommended to specify the programming language you are using as part of your prompt. - -## Chat Mode Scorecard Criteria - -When you visit Assistant's Chat Mode landing page, you will be presented with a scorecard comparing different aspects of each model. - -The Privacy column in the scorecard rates each model based on how it handles user data. The criteria are: - -- 1 circle: No data retention period specified. -- 2 circles: Data retained for more than 30 days before deletion. -- 3 circles: Data retained for 30 days or less before deletion. -- 4 circles: No data retained. - -None of the models train on user data. - -Check also [Kagi LLM Benchmarking Project](./llm-benchmark.md). -