Version v0.9.8

Version 0.9.8 adds support for new Gemini, Anthropic, OpenAI, Perplexity, and DeepSeek models, introduces LLM tool use/function calling, a redesign of gptel-menu, includes new customization hooks, dry-run options and refined settings, improvements to the rewrite feature and control of LLM “reasoning” content.

Breaking changes

• gemini-pro has been removed from the list of Gemini models, as this model is no longer supported by the Gemini API.

• Sending an active region in Org mode will now apply Org mode-specific rules to the text, such as branching context.

• The following obsolete variables and functions have been removed:

  • gptel-send-menu: Use gptel-menu instead.
  • gptel-host: Use gptel-make-openai instead.
  • gptel-playback: Use gptel-stream instead.
  • gptel--debug: Use gptel-log-level instead.

New models and backends

• Add support for several new Gemini models including gemini-2.0-flash, gemini-2.0-pro-exp and gemini-2.0-flash-thinking-exp, among others.

• Add support for the Anthropic model claude-3-7-sonnet-20250219, including its “reasoning” output.

• Add support for OpenAI’s o1, o3-mini and gpt-4.5-preview models.

• Add support for Perplexity. While gptel supported Perplexity in earlier releases by reusing its OpenAI support, there is now first class support for the Perplexity API, including citations.

• Add support for DeepSeek. While gptel supported DeepSeek in earlier releases by reusing its OpenAI support, there is now first class support for the DeepSeek API, including support for handling “reasoning” output.

New features and UI changes

• gptel-rewrite now supports iterating on responses.

• gptel supports the ability to simulate/dry-run requests so you can see exactly what will be sent. This payload preview can now be edited in place and the request continued.

• Directories can now be added to gptel’s global context. Doing so will add all files in the directory recursively.

• “Oneshot” settings: when using gptel’s Transient menus, request parameters, directives and tools can now be set for the next request only in addition to globally across the Emacs session and buffer-locally. This is useful for making one-off requests with different settings.

• gptel-mode can now be used in all modes derived from text-mode.

• gptel now tries to handle LLM responses that are in mixed Org/Markdown markup correctly.

• Add gptel-org-convert-response to toggle the automatic conversion of (possibly) Markdown-formatted LLM responses to Org markup where appropriate.

• You can now look up registered gptel backends using the gptel-get-backend function. This is intended to make scripting and configuring gptel easier. gptel-get-backend is a generalized variable so you can (un)set backends with setf.

• Tool use: gptel now supports LLM tool use, or function calling. Essentially you can equip the LLM with capabilities (such as filesystem access, web search, control of Emacs or introspection of Emacs’ state and more) that it can use to perform tasks for you. gptel runs these tools using argument values provided by the LLMs. This requires specifying tools, which are elisp functions with plain text descriptions of their arguments and results. gptel does not include any tools out of the box yet.

• You can look up registered gptel tools using the gptel-get-tool function. This is intended to make scripting and configuring gptel easier. gptel-get-tool is a generalized variable so you can (un)set tools with setf.

• New hooks for customization:

  • gptel-prompt-filter-hook runs in a temporary buffer containing the text to be sent, before the full query is created. It can be used for arbitrary text transformations to the source text.
  • gptel-post-request-hook runs after the request is sent, and (possibly) before any response is received. This is intended for preparatory/reset code.
  • gptel-post-rewrite-hook runs after a gptel-rewrite request is successfully and fully received.

• gptel-menu has been redesigned. It now shows a verbose description of what will be sent and where the output will go. This is intended to provide clarity on gptel’s default prompting behavior, as well as the effect of the various prompt/response redirection it provides. Incompatible combinations of options are now disallowed.

• The spacing between the end of the prompt and the beginning of the response in buffers is now customizable via gptel-response-separator, and can be any string.

• gptel-context-remove-all is now an interactive command.

• gptel now handles “reasoning” content produced by LLMs. Some LLMs include in their response a “thinking” or “reasoning” section. This text improves the quality of the LLM’s final output, but may not be interesting to you by itself. The new user option gptel-include-reasoning controls whether and how gptel displays this content.

• (Anthropic API only) Some LLM backends can cache content sent to it by gptel, so that only the newly included part of the text needs to be processed on subsequent conversation turns. This results in faster and significantly cheaper processing. The new user option gptel-cache can be used to specify caching preferences for prompts, the system message and/or tool definitions. This is supported only by the Anthropic API right now.

• (Org mode) Org property drawers are now stripped from the prompt text before sending queries. You can control this behavior or specify additional Org elements to ignore via gptel-org-ignore-elements. (For more complex pre-processing you can use gptel-prompt-filter-hook.)

Notable Bug fixes

• Fix response mix-up when running concurrent requests in Org mode buffers.
• gptel now works around an Org fontification bug where streaming responses in Org mode buffers sometimes caused source code blocks to remain unfontified.

Pull requests

• Fix bug where dry run updates status message by @daedsidog in #498
• README: add prefix string docs for gptel-org-branching-context by @erganemic in #499
• Fix typo by @gamboz in #505
• Possible bug fix 473 by @andrewdea in #490
• Support adding files in directories to context by @benthamite in #438
• Remove gptel-extensions package by @gavinhughes in #522
• Add windows platform check for the curl file size thresh by @AustinMooreT in #519
• chore: update gptel-openai.el by @eltociear in #535
• fix(private-gpt): passing system directive by @rosenstrauch in #531
• Add gemini-2.0-flash-thinking-exp by @jlcheng in #528
• Add support for OpenAI o1 model by @tillydray in #559
• Update examples by @Lyonsclay in #569
• Fix mark deactivation timing in gptel-rewrite by @ultronozm in #577
• Add gptel-perplexity backend by @pirminj in #581
• gptel: Use substitute-command-keys in context by @cashpw in #589
• Pass nil results through to model by @psionic-k in #596
• Parameters Required by @psionic-k in #600
• Variable number of newlines before responses by @psionic-k in #599
• Updates required Org version notice from 9.6.7 to 9.7 by @grettke in #602
• gptel-context: Use current buffer as default by @LemonBreezes in #595
• Improve robustness: gptel-org--link-standalone-p by @tillydray in #594
• gptel: Make gptel-model variable definition dynamic by @pabl0 in #606
• gptel-transient: Improve directive/system message editing by @pabl0 in #616
• gptel-transient: (really) unmark before editing crowdsourced prompt by @pabl0 in #618
• Fix read-only behaviour in gptel--edit-directive by @pabl0 in #622
• gptel-transient: Improve parsing of crowdsourced prompts CSV file by @pabl0 in #615
• Fix type error in read_buffer tool use example by @m-n in #632
• Small documentation tweaks by @pabl0 in #625
• gptel-transient: Ensure user enters a number when prompted by @pabl0 in #637
• Update gptel-gemini.el models by @braun-steven in #639
• gptel-transient: Fix face related bug by @pabl0 in #638
• gptel-transient: Use proper ellipsis for consistent UI by @pabl0 in #640
• gptel-transient: Report what is being sent from kill-ring by @pabl0 in #642
• gptel-openai-extras: reasoning and deep-research perplexity models by @markus1189 in #664
• gptel: Prefer LLM terminology over GPT by @pabl0 in #635
• Add support for gpt-4.5-preview by @benthamite in #673
• Fix typo in ...

Version v0.9.7

Version 0.9.7 adds dynamic directives, a better rewrite interface, streaming support to the gptel request API, and more flexible model/backend configuration

Breaking changes

gptel-rewrite-menu has been obsoleted. Use gptel-rewrite instead.

Backends

• Add support for OpenAI's o1-preview and o1-mini
• Add support for Anthropic's Claude 3.5 Haiku
• Add support for xAI (contributed by @WuuBoLin)
• Add support for Novita AI (contributed by @jasonhp)

Notable new features and UI changes

gptel's directives (see gptel-directives) can now be dynamic, and include more than the system message. You can "pre-fill" a conversation with canned user/LLM messages. Directives can now be functions that dynamically generate the system message and conversation history based on the current context. This paves the way for fully flexible task-specific templates, which the UI does not yet support in full. This design was suggested by @meain. (#375)

gptel's rewrite interface has been reworked. If using a streaming endpoint, the rewritten text is streamed in as a preview placed over the original. In all cases, clicking on the preview brings up a dispatch you can use to easily diff, ediff, merge, accept or reject the changes (4ae9c1b), and you can configure gptel to run one of these actions automatically. See the README for examples. This design was suggested by @meain. (#375)

gptel-abort, used to cancel requests in progress, now works across the board, including when not using Curl or with gptel-rewrite (7277c00).

The gptel-request API now explicitly supports streaming responses (7277c00), making it easy to write your own helpers or features with streaming support. The API also supports gptel-abort to stop and clean up responses.

You can now unset the system message -- different from setting it to an empty string. gptel will also automatically disable the system message when using models that don't support it (0a2c07a).

Support for including PDFs with requests to Anthropic models has been added. (These queries are cached, so you pay only 10% of the token cost of the PDF in follow-up queries.) Note that document support (PDFs etc) for Gemini models has been available since v0.9.5. (0f173ba, #459)

When defining a gptel model or backend, you can specify arbitrary parameters to be sent with each request. This includes the (many) API options across all APIs that gptel does not yet provide explicit support for (bcbbe67). This feature was suggested by @tillydray (#471).

New transient command option to easily remove all included context chunks (a844612), suggested by @metachip and @gavinhughes.

Bug fixes and other news

• Pressing RET on included files in the context inspector buffer now pops up the file correctly.
• API keys are stripped of whitespace before sending.
• Multiple UI, backend and prompt construction bugs have been fixed.

Pull requests

• Remove chatgpt-arcana from the list of alternatives by @CarlQLange in #431
• gptel-anthropic: Add upgraded Claude 3.5 Sonnet model by @benthamite in #436
• docs: add Novita AI in README by @jasonhp in #448
• gptel-curl: don't try converting CR-LF on Windows by @ssafar in #456
• README: Add support for xAI by @WuuBoLin in #466
• fix typo by @tillydray in #469
• fix(private-gpt): convert model name from symbol to string in request by @rosenstrauch in #470
• Update gptel--anthropic-models by @benthamite in #483
• Added a section tell new users how to select a backend explicitly by @blais in #467

New Contributors

• @CarlQLange made their first contribution in #431
• @jasonhp made their first contribution in #448
• @ssafar made their first contribution in #456
• @WuuBoLin made their first contribution in #466
• @tillydray made their first contribution in #469
• @rosenstrauch made their first contribution in #470
• @blais made their first contribution in #467

Full Changelog: v0.9.6...v0.9.7

Version v0.9.6

Version 0.9.6 is a bugfix release

New features and UI changes

gptel now displays more model details when switching the active LLM. This includes model descriptions, capabilities, context window sizes, input and output costs and the last time the model was updated by the provider. This feature was contributed by @benthamite.

Bug fixes

The last release included bugs that made gptel unusable with the Gemini backend. These have been fixed.

Contributors

• @benthamite made their first contribution in #417
• Add annotations next to model completion candidates by @benthamite in #420

Full Changelog: v0.9.5...v0.9.6

Version v0.9.5

Version 0.9.5 adds support for media/images, better rewrite interface, more models/backends

Breaking changes

The value of gptel-model, the LLM in use, is now expected to be a symbol, not a string. This change is necessary to support per-model capabilities like image support. As of this release, gptel deals with models specified as strings gracefully and issues a warning, but this will be an unhandled error in future releases.

Backends

• Add support for Deepseek (contributed by @jamorphy)
• Add support for Cerebras (#372, contributed by @bytegorilla)
• Add support for OpenAI's gpt-4o-mini
• Add support for Gemini's gemini-1.5-flash and gemini-1.5-pro
• Add support for Github Models (#386, contributed by @gs-101)
• Add support for system messages to Gemini models, as the API now supports them

New features

gptel now supports multi-modal models, i.e. models that can accept image or other binary input. Supported media files can be added to the context with gptel-add-file, or included as links in chat buffers. See the README for more information.

You can now specify more model metadata when defining gptel-models. This includes a description string, supported capabilities and a list of supported mime-types for multi-modal input. See the documentation of any of the backend-creation functions gptel-make-* for details.

UI changes

gptel's "rewrite" or "refactoring" interface has been redesigned. Rewriting is now a two step process, but stays out of your way and provides better control over how changes are applied: you can evaluate the suggested changes via a diff/ediff/merge conflict + smerge interface and apply them granularly. See the README for details.

Bug fixes and other news

• Fix incompatibilities with older Org versions (< 9.7).
• Fix prompt creation bugs in Org mode.
• Fix more bugs with setting transient switches buffer-locally in gptel-menu.

New Contributors, Acknowledgments

• @hron made their first contribution in #345
• @munen made their first contribution in #356
• @jdormit made their first contribution in #358
• @nfedyashev made their first contribution in #369
• @bytegorilla made their first contribution in #372
• @gs-101 made their first contribution in #386
• @xgqt made their first contribution in #398
• @jaeyeom made their first contribution in #412

Thanks to @daedsidog, @bradmont, @mateialexandru, @wlauppe and @nfedyashev for discussions and feedback.

Full Changelog: v0.9.0...v0.9.5

Version 0.9.0

Version 0.9.0 adds gptel-context and support for more models/backends

Backends

• Add support for OpenAI's gpt-4o (#313, contributed by @axelknock)
• Add support for Anthropic's claude-3-5-sonnet-20240620 (#331, contributed by @erwald).
• Add support for the PrivateGPT backend (#312, contributed by @Aquan1412).

New feature

gptel can now include arbitrary regions, buffers or files with requests. This is useful as background context for queries. For example, when you want to talk about the contents of code buffers/files in your project in a chat buffer. These additional contexts are "live" and not "snapshots", i.e. they are scanned at the time of the query.

This feature is available via the gptel-add command or from gptel's transient menu. See the README for more details.

This feature was contributed by @daedsidog.

UI changes

• Calling M-x gptel now asks the user to pick an existing chat buffer or name a new one. A suitable default name is chosen if the user leaves the field blank. The prefix-arg behavior of gptel has been removed for now.
• New option gptel-track-response in non-chat buffers to control whether gptel distinguishes between the user's prompts and LLM responses. (The two are always distinguished in dedicated chat buffers.) This option can also be set from the transient menu.

Bug fixes and other news

• gptel-org-branching-context now requires Org 9.67 or later.
• Fix bugs with one-shot Markdown to Org conversion.
• Fix bugs with setting transient switches buffer-locally in gptel-menu.

Version 0.8.6

Version 0.8.6 is a bugfix release

NonGNU ELPA

gptel is now available on NonGNU ELPA. This means it is directly installable via M-x package-install without the need to add MELPA to the list of package archives.

Backends

• Add support for OpenRouter (#282)
• Add support for Gemini 1.5 (#284)
• Add support for GPT 4 Turbo (#286)

Bug fixes

Several bugs have been fixed:

• gptel's status in the header line now updates when sending queries from the transient menu. (#293)
• More Org elements are supported in the Markdown to Org converter (#296)
• gptel now supports saving and resuming chats when using Ollama (#181)
• Ollama integration is now stateless, resolving a number of bugs (#270, #279)
• Fix Ollama response parsing errors (#179)

Version 0.8.5

Version 0.8.5 adds the following features:

gptel and Org mode

Additional features are now available when using gptel in Org mode:

• By default the context for a query is the contents of the buffer up to the cursor. You can set the context to be the lineage of the current Org heading by enabling gptel-org-branching-context. This makes each heading at a given level a different branch of the conversation.
• Limit the conversation context to an Org heading by setting gptel-org-set-topic. (This is an alternative to branching context.)
• Set the current gptel configuration (backend, model, system message, temperature and max tokens) as Org properties under the current heading with the command gptel-org-set-properties. When these properties are present, they override the buffer-specific or global settings. (#141)
• On-the-fly conversion of responses to Org mode has been smoothed out further. For best results, ask the LLM to respond only in Markdown if you are using Org mode.

UI

• The current system prompt is now shown (in truncated form) in the header-line in gptel chat buffers (#274).

Bug fixes

Several bugs have been fixed:

• (Anthropic) Attach additional directive correctly when interacting with Anthropic Claude models (#276)
• (Anthropic) Handle edge cases when parsing partial responses (#261)
• (All backends) Fix empty responses caused by json.el parsing errors when libjansson support is not available (#251, #264)
• (Ollama) Fix parsing errors caused by the libjansson transition (#255)
• The dry-run commands now perform an actual dry-run, i.e. everything except sending the queries (#276).
• The cursor can now be moved by functions in gptel-post-response-hook. (#269)

Version 0.8.0

Version 0.8.0 adds the following features:

Backends

• Support for the Anthropic API and Claude 3 models
• Support for Groq
• Updated OpenAI model list.

UI and configuration

There have been many improvements to the gptel's transient menu interface:

• From the menu, you can now attach an additional directive to the next query on top of the system message. This is useful for specific instructions that change with each query, like when refactoring code and regenerating or refining responses.
• Some introspection commands have been added to the menu: you can see exactly what will be sent (as a Lisp or JSON object) with the next query. To enable these commands in the menu, turn on gptel-log-level, which see.
• Various aspects of the menu have been tuned to be more efficient and cause less friction.
• Model and query parameters (including the system message) are now global variables by default. This is to make it easier to work at a project level without having to set them in each buffer. You can continue to set them at a buffer-local level (the previous default) using a switch in the menu.

Other

• gptel now uses libjansson if Emacs is compiled with support for it. This makes json parsing about 3x faster. (LLM responses are typically not large, so there is only a modest increase in parsing speed and Emacs' responsiveness when using gptel.)
• Org mode output when streaming responses is much improved, with most edge cases resolved.

Deprecation notice

• The dedicated "refactor" or "rewrite" menus are deprecated and will be removed in the next major release. Note that all of their functionality (including ediff-ing) is now available from the main gptel menu.

Version 0.7.0

Version 0.7.0 adds the following features:

Backends

• Support for Perplexity.ai (contributed by @dbactual)
• Updated OpenAI model list.
• Better support for Gemini
• Support for arbitrary curl arguments (for HTTP requests) that gptel does not provide customization options for (contributed by @r0man)

UI and configuration

• Response regeneration and history: You can now regenerate a response at point and cycle through past versions of the response at point. These are accessible from the transient menu when the point is over a response.
• Customizable display-buffer action to choose where gptel chat windows are placed when chat buffers are created with M-x gptel.

Version 0.6.5

Version 0.6.5 adds the following features:

Backends

• Support for the Kagi summarizer engine(s).
• Support for Together.ai and Anyscale.
• Updated OpenAI model list.

UI and configuration

• gptel handles the case of trying to use unsupported models for a backend (contributed by @joaotavora).
• Easier configuration for OpenAI backends (header does not need to be specified)
• Redirect responses to any buffer, not just gptel chat sessions.
• Optional logging of all requests/response data.

Bug fixes

• Persist multi-line directives in saved files correctly.
• Org-conversion bug fixes.

Many more minor bug fixes.