Eden supports OpenAI, Anthropic, Deepseek, Perplexity, and X.ai APIs. For reasoning models, you can choose whether or not to display the reasoning. For Perplexity and OpenAI web search, while citations are inlined, you can also list them in a dedicated buffer.

Eden’s interface is simple:

• You want to ask something to ChatGPT? Call eden command, enter your prompt, press C-c C-c and you’re done.

  

• You want to integrate the response in your org-mode notes? Just copy/paste it.
• You want to continue a conversation saved in your org-mode notes? On the request at point, call eden with the C-u prefix argument which opens a transient menu, then press c to to continue the conversation whose last request is the request at point.

  

• You want to continue a conversation from a previous request? Simple! In the prompt buffer, navigate through the prompt history using M-p and M-n to find the desired request, open the menu calling eden and press c to continue the conversation from that request.

  

• You want to switch the API and the model? Just call eden in the prompt buffer, then press a to change the API and m to change the model.

  

• You want to read the reasoning behind the response? Great! In the prompt buffer, call eden and press i to toggle the reasoning. Send your request and get the reasoning in its own section.

  

• You want to list Perplexity citations? Fine! On the request at point, call eden with the C-u prefix argument, then press C.

  

Requests are independent by default and are appended to the *eden[requests]* buffer. To link them together, you must be in a conversation; in that case, they are appended to unique buffers named like this: *eden[<conversation's name>]*.

Note that all requests are stored in the eden-dir directory, which defaults to the eden subdirectory within your user-emacs-directory.

Also note that the prompt buffer *eden* uses eden-mode, which is derived from org-mode. This means that everything you can do in org-mode is also available in the prompt buffer, except for the following key bindings: C-c C-c, M-p, M-n, C-M-p, and C-M-n.

1. Ensure the following utilities are installed and present in one of your exec-path directories:
   • curl
   • uuidgen
   • pandoc

   You can ask ChatGPT to help you with this (replace distribution with your distribution be Mac, Ubuntu, etc.):

   How to install the utilities curl, uuidgen, and pandoc on <distribution>?
       

2. Add the directory containing eden.el to your load-path and require the Eden package by adding the following lines to your init file, ensuring to replace /path/to/eden/ with the appropriate directory:

   (add-to-list 'load-path "/path/to/eden/")
   (require 'eden)
       

   Or if you’re using straight:

   (straight-use-package
    '(eden :type git :host github :repo "tonyaldon/eden"))
       

3. Store your OpenAI API key in either the ~/.authinfo.gpg file (encrypted with gpg) or the ~/.authinfo file (plaintext):
   • After funding your OpenAI account ($5.00 is enough to get started), create an OpenAI API key visiting https://platform.openai.com/api-keys. You can check my video if you’ve never done this before.
   • Add the API key in the selected file as follows:

     machine openai password <openai-api-key>
             

     where <openai-api-key> is your API key.

   • Restart Emacs to apply this change.
4. Call the command eden to switch to *eden* prompt buffer,
5. Enter your prompt,
6. Press C-c C-c to send your prompt to OpenAI API,
7. Finally, the response will asynchronously show up in a dedicated buffer upon receipt.

To use the APIs from OpenAI, Anthropic, Deepseek, Perplexity, or X.ai you need to store their API key in either the ~/.authinfo.gpg file (encrypted with gpg) or the ~/.authinfo file (plaintext) as follow:

machine openai password <openai-api-key>
machine anthropic password <anthropic-api-key>
machine deepseek password <deepseek-api-key>
machine perplexity password <perplexity-api-key>
machine x.ai password <x.ai-api-key>

You can create new API keys at the following links:

• https://platform.openai.com/api-keys
• https://console.anthropic.com/settings/keys
• https://platform.deepseek.com/api_keys
• https://perplexity.ai/account/api/keys
• https://console.x.ai

Eden focuses on conversations without enforcing them; defaulting to independent requests, it makes starting new conversations or continuing from previous ones easy!

There are several ways to engage in a conversation while in the prompt buffer:

1. To start a new conversation, call the eden command, press n, and enter a title. This creates a new empty conversation, setting it as the current one for all new requests.
2. To start a conversation from the current request in history (excluding earlier exchanges), navigate through the history using M-p and M-n to find the desired request. Then, call the eden command, press s, and enter a title. This creates a new conversation that already include one exchange.
3. To continue an existing conversation call the eden command, press c, and enter a title. This will include all previous exchanges of the current request in history.

You can pause the current conversation by calling eden and pressing SPC. Subsequent requests will then be independent again.

When you are in a conversation, the name of the conversation appears in the mode line of the prompt buffer, enclosed in brackets.

Note that conversation titles and IDs are not stored; they only exist during your Emacs session. However, you can retrieve any conversation later either by saving it in your notes (with its UUID) or navigating the prompt history with M-p and M-n.

All requests are stored in the eden-dir directory, which defaults to the eden subdirectory within your user-emacs-directory, providing a range of benefits:

1. Requests are always preserved, ensuring you can retrieve them at any time.
2. With the request’s UUID, you can track down the associated request and check details like the API, model, system prompt, and timestamp.
3. Should an error occur during processing, the corresponding error.json file can be consulted for troubleshooting.
4. You can start or continue a conversation from any existing request (a feature known as “branching”):
   • Either from a request at point in your notes,
   • Or navigating through history in the prompt buffer using M-p and M-n to find the desired request, opening the menu with eden, and pressing c to continue the conversation or s to start a new conversation from the request.
5. All data is stored in JSON (or text format), facilitating integration with other software for further analysis.

In the prompt buffer, you can call eden and press C to show the current configuration.

From the menu provided by eden, you can modify the current configuration doing the following

• Press a to set the current API (eden-api-set),
• Press m to set the model for the current API (eden-model-set),
• Press t to set the temperature (eden-temperature-set),
• Press i to include reasoning information (eden-include-reasoning-toggle),
• Press o to access another menu with more options (eden-more-options-menu),

or the following for modifying the system message:

• Press S to add a system message (eden-system-message-add),
• Press u to update the system message (eden-system-message-update),
• Press p to pick a system message (eden-system-message-set),
• Press r to reset the system message (eden-system-message-reset).

Each time you quit Eden’s menu, the current profile which includes

• the API,
• the request directory,
• the model,
• if you include the reasoning,
• the temperature,
• the current conversation if any, and,
• the system message if any
• the instructions appended to the system message if any

is pushed to an history that you can navigate from the prompt buffer with:

• C-M-p bound to eden-profile-previous and
• C-M-n bound to eden-profile-next.

The complete list of user variables you may want to adjust includes:

• eden-api
• eden-apis
• eden-model
• eden-temperature
• eden-system-message
• eden-system-message-append
• eden-system-messages
• eden-system-message->developer-for-models
• eden-dir
• eden-anthropic-max-tokens
• eden-anthropic-thinking-budget-tokens
• eden-web-search-context-size
• eden-org-property-date
• eden-org-property-model
• eden-org-property-req
• eden-pops-up-upon-receipt
• eden-include-reasoning
• eden-prompt-buffer-name

For more information on these variables, consult their documentation in the *Help* buffer using describe-variable command, bound by default to C-h v.

See “Alternatives” section of gptel README for a comprehensive list of Emacs clients for LLMs not limited to OpenAI.

No.

I don’t like it.

Streaming the response forces me to read it immediately and linearly.

That’s not how I read. I often start from the end and go backward, picking out only the pieces I’m interested in. If I need a more profound understanding of the answer, I might then read it linearly to make sure I don’t miss anything.

And if I have to read the entire text of each response, I’ll get exhausted too quickly. My processing power can’t keep up with the production rate of LLMs. I have to choose wisely what I read and what I don’t.

You might say, “Nobody is forcing you to read it this way; you can just wait until the end.” That’s true! But in that case, why bother implementing streaming at all?