diff --git a/NEWS b/NEWS
index cb7e2975..3e9ca674 100644
--- a/NEWS
+++ b/NEWS
@@ -7,8 +7,33 @@
- Add support for ~gemini-2.5-pro~, ~gemini-2.5-flash~,
~gemini-2.5-flash-lite-preview-06-17~.
+- Add support for Open WebUI. Open WebUI provides an
+ OpenAI-compatible API, so the "support" is just a new section of the
+ README with instructions.
+
** New features and UI changes
+- When including a file in the context, the abbreviated full path of
+ the file is included is now included instead of the basename.
+ Specifically, =/home/user/path/to/file= is included as
+ =~/path/to/file=. This is to provide additional context for LLM
+ actions, including tool-use in subsequent conversation turns. This
+ applies to context included via ~gptel-add~ or as a link in a
+ buffer.
+
+- Structured output support: ~gptel-request~ can now take an optional
+ schema argument to constrain LLM output to the specified JSON
+ schema. The JSON schema can be provided as a serialized JSON string
+ or as an elisp object (a nested plist). This feature works with all major
+ backends: OpenAI, Anthropic, Gemini, llama-cpp and Ollama. It is
+ presently supported by some but not all "OpenAI-compatible API"
+ providers. Note that this is only available via the ~gptel-request~
+ API, and currently unsupported by ~gptel-send~.
+
+- From the dry-run inspector buffer, you can now copy the Curl command
+ for the request. Like when continuing the query, the request is
+ constructed from the contents of the buffer, which is editable.
+
- gptel now handles Ollama models that return both reasoning content
and tool calls in a single request.
diff --git a/README.org b/README.org
index 5cc6c78a..349cdaae 100644
--- a/README.org
+++ b/README.org
@@ -14,6 +14,7 @@ gptel is a simple Large Language Model chat client for Emacs, with support for m
| Anthropic (Claude) | ✓ | [[https://www.anthropic.com/api][API key]] |
| Gemini | ✓ | [[https://makersuite.google.com/app/apikey][API key]] |
| Ollama | ✓ | [[https://ollama.ai/][Ollama running locally]] |
+| Open WebUI | ✓ | [[https://openwebui.com/][Open WebUI running locally]] |
| Llama.cpp | ✓ | [[https://github.com/ggml-org/llama.cpp/tree/master/tools/server#quick-start][Llama.cpp running locally]] |
| Llamafile | ✓ | [[https://github.com/Mozilla-Ocho/llamafile#quickstart][Local Llamafile server]] |
| GPT4All | ✓ | [[https://gpt4all.io/index.html][GPT4All running locally]] |
@@ -24,6 +25,7 @@ gptel is a simple Large Language Model chat client for Emacs, with support for m
| Mistral Le Chat | ✓ | [[https://console.mistral.ai/api-keys][API key]] |
| Perplexity | ✓ | [[https://docs.perplexity.ai/docs/getting-started][API key]] |
| OpenRouter | ✓ | [[https://openrouter.ai/keys][API key]] |
+| AI/ML API | ✓ | [[https://aimlapi.com/app/?utm_source=gptel&utm_medium=github&utm_campaign=integration][API key]] |
| together.ai | ✓ | [[https://api.together.xyz/settings/api-keys][API key]] |
| Anyscale | ✓ | [[https://docs.endpoints.anyscale.com/][API key]] |
| PrivateGPT | ✓ | [[https://github.com/zylon-ai/private-gpt#-documentation][PrivateGPT running locally]] |
@@ -101,6 +103,7 @@ gptel uses Curl if available, but falls back to the built-in url-retrieve to wor
- [[#azure][Azure]]
- [[#gpt4all][GPT4All]]
- [[#ollama][Ollama]]
+ - [[#open-webui][Open WebUI]]
- [[#gemini][Gemini]]
- [[#llamacpp-or-llamafile][Llama.cpp or Llamafile]]
- [[#kagi-fastgpt--summarizer][Kagi (FastGPT & Summarizer)]]
@@ -118,6 +121,7 @@ gptel uses Curl if available, but falls back to the built-in url-retrieve to wor
- [[#github-models][Github Models]]
- [[#novita-ai][Novita AI]]
- [[#xai][xAI]]
+ - [[#aiml-api][AI/ML API]]
- [[#github-copilotchat][Github CopilotChat]]
- [[#aws-bedrock][AWS Bedrock]]
- [[#usage][Usage]]
@@ -333,6 +337,62 @@ The above code makes the backend available to select. If you want it to be the
#+html:
+#+html:
+**** Open WebUI
+#+html:
+
+[[https://openwebui.com/][Open WebUI]] is an open source, self-hosted system which provides a multi-user web chat interface and an API endpoint for accessing LLMs, especially LLMs running locally on inference servers like Ollama.
+
+Because it presents an OpenAI-compatible endpoint, you use ~gptel-make-openai~ to register it as a backend.
+
+For instance, you can use this form to register a backend for a local instance of Open Web UI served via http on port 3000:
+
+#+begin_src emacs-lisp
+(gptel-make-openai "OpenWebUI"
+ :host "localhost:3000"
+ :protocol "http"
+ :key "KEY_FOR_ACCESSING_OPENWEBUI"
+ :endpoint "/api/chat/completions"
+ :stream t
+ :models '("gemma3n:latest"))
+#+end_src
+
+Or if you are running Open Web UI on another host on your local network (~box.local~), serving via https with self-signed certificates, this will work:
+
+#+begin_src emacs-lisp
+(gptel-make-openai "OpenWebUI"
+ :host "box.local"
+ :curl-args '("--insecure") ; needed for self-signed certs
+ :key "KEY_FOR_ACCESSING_OPENWEBUI"
+ :endpoint "/api/chat/completions"
+ :stream t
+ :models '("gemma3n:latest"))
+#+end_src
+
+To find your API key in Open WebUI, click the user name in the bottom left, Settings, Account, and then Show by API Keys section.
+
+Refer to the documentation of =gptel-make-openai= for more configuration options.
+
+You can pick this backend from the menu when using gptel (see [[#usage][Usage]])
+
+***** (Optional) Set as the default gptel backend
+
+The above code makes the backend available to select. If you want it to be the default backend for gptel, you can set this as the value of =gptel-backend=. Use this instead of the above.
+#+begin_src emacs-lisp
+;; OPTIONAL configuration
+(setq
+ gptel-model "gemma3n:latest"
+ gptel-backend (gptel-make-openai "OpenWebUI"
+ :host "localhost:3000"
+ :protocol "http"
+ :key "KEY_FOR_ACCESSING_OPENWEBUI"
+ :endpoint "/api/chat/completions"
+ :stream t
+ :models '("gemma3n:latest")))
+#+end_src
+
+#+html:
**** Gemini
#+html:
@@ -924,6 +984,41 @@ The above code makes the backend available to select. If you want it to be the
#+html:
+**** AI/ML API
+#+html:
+
+AI/ML API provides 300+ AI models including Deepseek, Gemini, ChatGPT. The models run at enterprise-grade rate limits and uptimes.
+
+Register a backend with
+#+begin_src emacs-lisp
+;; AI/ML API offers an OpenAI compatible API
+(gptel-make-openai "AI/ML API" ;Any name you want
+ :host "api.aimlapi.com"
+ :endpoint "/v1/chat/completions"
+ :stream t
+ :key "your-api-key" ;can be a function that returns the key
+ :models '(deepseek-chat gemini-pro gpt-4o))
+#+end_src
+
+You can pick this backend from the menu when using gptel (see [[#usage][Usage]]).
+
+***** (Optional) Set as the default gptel backend
+
+The above code makes the backend available to select. If you want it to be the default backend for gptel, you can set this as the value of =gptel-backend=. Use this instead of the above.
+#+begin_src emacs-lisp
+;; OPTIONAL configuration
+(setq gptel-model 'gpt-4o
+ gptel-backend
+ (gptel-make-openai "AI/ML API"
+ :host "api.aimlapi.com"
+ :endpoint "/v1/chat/completions"
+ :stream t
+ :key "your-api-key"
+ :models '(deepseek-chat gemini-pro gpt-4o)))
+#+end_src
+
+#+html:
**** Github CopilotChat
#+html:
@@ -1526,13 +1621,15 @@ Other Emacs clients for LLMs prescribe the format of the interaction (a comint s
all)
#+end_src
-|----------------------+--------------------------------------------------------------------|
-| *Connection options* | |
-|----------------------+--------------------------------------------------------------------|
-| =gptel-use-curl= | Use Curl (default), fallback to Emacs' built-in =url=. |
-| =gptel-proxy= | Proxy server for requests, passed to curl via =--proxy=. |
-| =gptel-api-key= | Variable/function that returns the API key for the active backend. |
-|----------------------+--------------------------------------------------------------------|
+|-------------------------+--------------------------------------------------------------------|
+| *Connection options* | |
+|-------------------------+--------------------------------------------------------------------|
+| =gptel-use-curl= | Use Curl? (default), fallback to Emacs' built-in =url=. |
+| | You can also specify the Curl path here. |
+| =gptel-proxy= | Proxy server for requests, passed to curl via =--proxy=. |
+| =gptel-curl-extra-args= | Extra arguments passed to Curl. |
+| =gptel-api-key= | Variable/function that returns the API key for the active backend. |
+|-------------------------+--------------------------------------------------------------------|
|-----------------------+---------------------------------------------------------|
| *LLM request options* | /(Note: not supported uniformly across LLMs)/ |
@@ -1677,10 +1774,17 @@ There are several more: [[https://github.com/iwahbe/chat.el][chat.el]], [[https:
gptel is a general-purpose package for chat and ad-hoc LLM interaction. The following packages use gptel to provide additional or specialized functionality:
- [[https://github.com/karthink/gptel-quick][gptel-quick]]: Quickly look up the region or text at point.
+- [[https://github.com/jwiegley/gptel-prompts][gptel-prompts]]: System prompt manager for gptel.
+- [[https://github.com/dolmens/gptel-aibo/][gptel-aibo]]: A writing assistant system built on top of gptel.
+- [[https://github.com/kmontag/macher][Macher]]: Project-aware multi-file LLM editing for Emacs, based on gptel.
- [[https://github.com/daedsidog/evedel][Evedel]]: Instructed LLM Programmer/Assistant.
- [[https://github.com/lanceberge/elysium][Elysium]]: Automatically apply AI-generated changes as you code.
+- [[https://github.com/jwiegley/ob-gptel][ob-gptel]]: Org-babel backend for running gptel queries.
+- [[https://github.com/JDNdeveloper/gptel-autocomplete][gptel-autocomplete]]: Inline completions using gptel.
- [[https://github.com/kamushadenes/ai-blog.el][ai-blog.el]]: Streamline generation of blog posts in Hugo.
-- [[https://github.com/douo/magit-gptcommit][magit-gptcommit]]: Generate Commit Messages within magit-status Buffer using gptel.
+- [[https://github.com/lakkiy/gptel-commit][gptel-commit]]: Generate commit messages using gptel.
+- [[https://github.com/douo/magit-gptcommit][magit-gptcommit]]: Generate commit messages within magit-status Buffer using gptel.
+- [[https://github.com/ragnard/gptel-magit/][gptel-magit]]: Generate commit messages for magit using gptel.
- [[https://github.com/armindarvish/consult-omni][consult-omni]]: Versatile multi-source search package. It includes gptel as one of its many sources.
- [[https://github.com/ultronozm/ai-org-chat.el][ai-org-chat]]: Provides branching conversations in Org buffers using gptel. (Note that gptel includes this feature as well (see =gptel-org-branching-context=), but requires a recent version of Org mode 9.7 or later to be installed.)
- [[https://github.com/rob137/Corsair][Corsair]]: Helps gather text to populate LLM prompts for gptel.
diff --git a/gptel-anthropic.el b/gptel-anthropic.el
index 48714dc0..181e79f3 100644
--- a/gptel-anthropic.el
+++ b/gptel-anthropic.el
@@ -230,6 +230,13 @@ Mutate state INFO with response metadata."
(gptel--model-capable-p 'cache))
(nconc (aref tools-array (1- (length tools-array)))
'(:cache_control (:type "ephemeral")))))))
+ (when gptel--schema
+ (plist-put prompts-plist :tools
+ (vconcat
+ (list (gptel--parse-schema backend gptel--schema))
+ (plist-get prompts-plist :tools)))
+ (plist-put prompts-plist :tool_choice
+ `(:type "tool" :name ,gptel--ersatz-json-tool)))
;; Merge request params with model and backend params.
(gptel--merge-plists
prompts-plist
@@ -237,6 +244,15 @@ Mutate state INFO with response metadata."
(gptel-backend-request-params gptel-backend)
(gptel--model-request-params gptel-model))))
+(cl-defmethod gptel--parse-schema ((_backend gptel-anthropic) schema)
+ ;; Unlike the other backends, Anthropic generates JSON using a tool call. We
+ ;; write the tool here, meant to be added to :tools.
+ (list
+ :name "response_json"
+ :description "Record JSON output according to user prompt"
+ :input_schema (gptel--preprocess-schema
+ (gptel--dispatch-schema-type schema))))
+
(cl-defmethod gptel--parse-tools ((_backend gptel-anthropic) tools)
"Parse TOOLS to the Anthropic API tool definition spec.
diff --git a/gptel-curl.el b/gptel-curl.el
index 018ff07a..8873c0bc 100644
--- a/gptel-curl.el
+++ b/gptel-curl.el
@@ -112,8 +112,6 @@ REQUEST-DATA is the data to send, TOKEN is a unique identifier."
collect (format "-H%s: %s" key val))
(list url))))
-;;TODO: The :transformer argument here is an alternate implementation of
-;;`gptel-response-filter-functions'. The two need to be unified.
;;;###autoload
(defun gptel-curl-get-response (fsm)
"Fetch response to prompt in state FSM from the LLM using Curl.
@@ -283,6 +281,9 @@ Optional RAW disables text properties and transformation."
(set-marker-insertion-type tracking-marker t)
(plist-put info :tracking-marker tracking-marker))
(goto-char tracking-marker)
+ (when (plist-get info :last-was-tool-result)
+ (insert gptel-response-separator)
+ (plist-put info :last-was-tool-result nil))
(unless raw
(when transformer
(setq response (funcall transformer response)))
@@ -297,7 +298,8 @@ Optional RAW disables text properties and transformation."
(`(tool-call . ,tool-calls)
(gptel--display-tool-calls tool-calls info))
(`(tool-result . ,tool-results)
- (gptel--display-tool-results tool-results info))))
+ (gptel--display-tool-results tool-results info)
+ (plist-put info :last-was-tool-result t))))
(defun gptel-curl--stream-filter (process output)
(let* ((fsm (car (alist-get process gptel--request-alist)))
@@ -366,7 +368,7 @@ Optional RAW disables text properties and transformation."
(progn (setq response (cons 'reasoning response))
(plist-put proc-info :reasoning-block 'in))
(plist-put proc-info :reasoning-block 'done)))
- ((length> response 0)
+ ((and (not (eq reasoning-block t)) (length> response 0))
(if-let* ((idx (string-match-p "" response)))
(progn
(funcall callback
@@ -374,10 +376,9 @@ Optional RAW disables text properties and transformation."
(string-trim-left
(substring response nil (+ idx 8))))
proc-info)
- ;; Signal end of reasoning stream
- (funcall callback '(reasoning . t) proc-info)
- (setq response (substring response (+ idx 8)))
- (plist-put proc-info :reasoning-block 'done))
+ (setq reasoning-block t) ;Signal end of reasoning stream
+ (plist-put proc-info :reasoning-block t)
+ (setq response (substring response (+ idx 8))))
(setq response (cons 'reasoning response)))))
(when (eq reasoning-block t) ;End of reasoning block
(funcall callback '(reasoning . t) proc-info)
diff --git a/gptel-gemini.el b/gptel-gemini.el
index f29e3232..e58863b5 100644
--- a/gptel-gemini.el
+++ b/gptel-gemini.el
@@ -114,15 +114,15 @@ list."
(cl-defmethod gptel--request-data ((backend gptel-gemini) prompts)
"JSON encode PROMPTS for sending to Gemini."
(let ((prompts-plist
- `(:contents [,@prompts]
- :safetySettings [(:category "HARM_CATEGORY_HARASSMENT"
- :threshold "BLOCK_NONE")
- (:category "HARM_CATEGORY_SEXUALLY_EXPLICIT"
- :threshold "BLOCK_NONE")
- (:category "HARM_CATEGORY_DANGEROUS_CONTENT"
- :threshold "BLOCK_NONE")
- (:category "HARM_CATEGORY_HATE_SPEECH"
- :threshold "BLOCK_NONE")]))
+ (list :contents (vconcat prompts)
+ :safetySettings [(:category "HARM_CATEGORY_HARASSMENT"
+ :threshold "BLOCK_NONE")
+ (:category "HARM_CATEGORY_SEXUALLY_EXPLICIT"
+ :threshold "BLOCK_NONE")
+ (:category "HARM_CATEGORY_DANGEROUS_CONTENT"
+ :threshold "BLOCK_NONE")
+ (:category "HARM_CATEGORY_HATE_SPEECH"
+ :threshold "BLOCK_NONE")]))
params)
(if gptel--system-message
(plist-put prompts-plist :system_instruction
@@ -145,6 +145,9 @@ list."
(when gptel-include-reasoning
(setq params
(plist-put params :thinkingConfig '(:includeThoughts t))))
+ (when gptel--schema
+ (setq params (nconc params (gptel--gemini-filter-schema
+ (gptel--parse-schema backend gptel--schema)))))
(when params
(plist-put prompts-plist
:generationConfig params))
@@ -155,6 +158,11 @@ list."
(gptel-backend-request-params gptel-backend)
(gptel--model-request-params gptel-model))))
+(cl-defmethod gptel--parse-schema ((_backend gptel-gemini) schema)
+ (list :responseMimeType "application/json"
+ :responseSchema (gptel--preprocess-schema
+ (gptel--dispatch-schema-type schema))))
+
(defun gptel--gemini-filter-schema (schema)
"Destructively filter unsupported attributes from SCHEMA.
@@ -471,7 +479,7 @@ files in the context."
:output-cost 10.00 ; 15 for >200k tokens
:cutoff-date "2025-01")
(gemini-2.5-flash-preview-04-17
- :description "Best model in terms of price-performance, offering well-rounded capabilities"
+ :description "DEPRECATED: Please use gemini-2.5-flash instead."
:capabilities (tool-use json media)
:mime-types ("image/png" "image/jpeg" "image/webp" "image/heic" "image/heif"
"application/pdf" "text/plain" "text/csv" "text/html")
@@ -480,7 +488,7 @@ files in the context."
:output-cost 0.60 ; 3.50 for thinking
:cutoff-date "2025-01")
(gemini-2.5-flash-preview-05-20
- :description "Best model in terms of price-performance, offering well-rounded capabilities"
+ :description "DEPRECATED: Please use gemini-2.5-flash instead."
:capabilities (tool-use json media)
:mime-types ("image/png" "image/jpeg" "image/webp" "image/heic" "image/heif"
"application/pdf" "text/plain" "text/csv" "text/html")
@@ -488,6 +496,17 @@ files in the context."
:input-cost 0.15
:output-cost 0.60 ; 3.50 for thinking
:cutoff-date "2025-01")
+ (gemini-2.5-flash
+ :description "Best model in terms of price-performance, offering well-rounded capabilities"
+ :capabilities (tool-use json media audio video)
+ :mime-types ("image/png" "image/jpeg" "image/webp" "image/heic" "image/heif"
+ "application/pdf" "text/plain" "text/csv" "text/html"
+ "audio/mpeg" "audio/wav" "audio/ogg" "audio/flac" "audio/aac" "audio/mp3"
+ "video/mp4" "video/mpeg" "video/avi" "video/quicktime" "video/webm")
+ :context-window 1024
+ :input-cost 0.15
+ :output-cost 0.60
+ :cutoff-date "2025-01")
(gemini-2.5-pro-preview-05-06
:description "Previously most powerful thinking model with state-of-the-art performance"
:capabilities (tool-use json media)
@@ -525,13 +544,27 @@ files in the context."
:output-cost 2.50
:cutoff-date "2025-01")
(gemini-2.5-pro
- :description "State-of-the-art thinking model, capable of reasoning over complex problems"
- :capabilities (tool-use json media)
+ :description "Most powerful Gemini thinking model with state-of-the-art performance"
+ :capabilities (tool-use json media audio video)
:mime-types ("image/png" "image/jpeg" "image/webp" "image/heic" "image/heif"
- "application/pdf" "text/plain" "text/csv" "text/html")
- :context-window 1048 ; 65536 output token limit
- :input-cost 1.25 ; 2.50 for >200k tokens
- :output-cost 10.00 ; 15 for >200k tokens
+ "application/pdf" "text/plain" "text/csv" "text/html"
+ "audio/mpeg" "audio/wav" "audio/ogg" "audio/flac" "audio/aac" "audio/mp3"
+ "video/mp4" "video/mpeg" "video/avi" "video/quicktime" "video/webm")
+ :context-window 1048 ; 65536 output token limit
+ :input-cost 1.25 ; 2.50 for >200k tokens
+ :output-cost 10.00 ; 15 for >200k tokens
+ :cutoff-date "2025-01")
+ (gemini-2.5-flash-lite-preview-06-17
+ :description "Fastest and cheapest 2.5 model, for high-volume, latency-sensitive tasks."
+ :capabilities (tool-use json media audio video)
+ :mime-types ("image/png" "image/jpeg" "image/webp" "application/pdf" "text/plain"
+ "audio/x-aac" "audio/flac" "audio/mp3" "audio/m4a" "audio/mpeg"
+ "audio/mpga" "audio/mp4" "audio/opus" "audio/pcm" "audio/wav" "audio/webm"
+ "video/x-flv" "video/quicktime" "video/mpeg" "video/mp4"
+ "video/webm" "video/wmv" "video/3gpp")
+ :context-window 1048
+ :input-cost 0.075
+ :output-cost 0.30
:cutoff-date "2025-01")
(gemini-2.0-flash-thinking-exp
:description "DEPRECATED: Please use gemini-2.0-flash-thinking-exp-01-21 instead."
diff --git a/gptel-ollama.el b/gptel-ollama.el
index 26b4b75d..1c5a07af 100644
--- a/gptel-ollama.el
+++ b/gptel-ollama.el
@@ -101,7 +101,10 @@ Store response metadata in state INFO."
(gptel--merge-plists
`(:model ,(gptel--model-name gptel-model)
:messages [,@prompts]
- :stream ,(or gptel-stream :json-false))
+ :stream ,(or gptel-stream :json-false)
+ ,@(and gptel--schema
+ `(:format ,(gptel--preprocess-schema
+ (gptel--dispatch-schema-type gptel--schema)))))
gptel--request-params
(gptel-backend-request-params gptel-backend)
(gptel--model-request-params gptel-model)))
diff --git a/gptel-openai-extras.el b/gptel-openai-extras.el
index 9f2c8caa..f2c3d7b0 100644
--- a/gptel-openai-extras.el
+++ b/gptel-openai-extras.el
@@ -294,11 +294,11 @@ parameters."
;; and reset by the stream filter.
(plist-put info :reasoning
(concat (plist-get info :reasoning) reasoning))
- (when-let* ((content (plist-get delta :content))
- ((not (eq content :null))))
- (if (eq (plist-get info :reasoning-block) 'in) ;Check if in reasoning block
- (plist-put info :reasoning-block t) ;End of streaming reasoning block
- (plist-put info :reasoning-block 'done)) ;Not using a reasoning model
+ ;; Done with reasoning if we get non-empty content
+ (when-let* (((plist-member info :reasoning)) ;Is this a reasoning model?
+ (content (plist-get delta :content)) ;Started receiving text content?
+ ((not (or (eq content :null) (string-empty-p content)))))
+ (plist-put info :reasoning-block t) ;Signal end of reasoning block
(throw 'done t)))))))))))
(cl-defmethod gptel--parse-response :before ((_backend gptel-deepseek) response info)
diff --git a/gptel-openai.el b/gptel-openai.el
index 241a9b56..f4060406 100644
--- a/gptel-openai.el
+++ b/gptel-openai.el
@@ -40,6 +40,7 @@
(defvar gptel-track-media)
(defvar gptel-use-tools)
(defvar gptel-tools)
+(defvar gptel--schema)
(declare-function gptel-context--collect-media "gptel-context")
(declare-function gptel--base64-encode "gptel")
(declare-function gptel--trim-prefixes "gptel")
@@ -47,6 +48,7 @@
(declare-function gptel--model-capable-p "gptel")
(declare-function gptel--model-name "gptel")
(declare-function gptel--get-api-key "gptel")
+(declare-function gptel--insert-file-string "gptel")
(declare-function prop-match-value "text-property-search")
(declare-function text-property-search-backward "text-property-search")
(declare-function json-read "json")
@@ -57,6 +59,7 @@
(declare-function gptel-context--wrap "gptel-context")
(declare-function gptel--inject-prompt "gptel")
(declare-function gptel--parse-tools "gptel")
+(declare-function gptel--parse-schema "gptel")
;; JSON conversion semantics used by gptel
;; empty object "{}" => empty list '() == nil
@@ -216,7 +219,9 @@ information if the stream contains it."
;; No text content, so look for tool calls
(when-let* ((tool-call (map-nested-elt delta '(:tool_calls 0)))
(func (plist-get tool-call :function)))
- (if (plist-get func :name) ;new tool block begins
+ (if (and (plist-get func :name)
+ ;; TEMP: This check is for litellm compatibility, should be removed
+ (not (equal (plist-get func :name) "null"))) ; new tool block begins
(progn
(when-let* ((partial (plist-get info :partial_json)))
(let* ((prev-tool-call (car (plist-get info :tool-use)))
@@ -306,6 +311,9 @@ Mutate state INFO with response metadata."
(plist-put prompts-plist
(if reasoning-model-p :max_completion_tokens :max_tokens)
gptel-max-tokens))
+ (when gptel--schema
+ (plist-put prompts-plist
+ :response_format (gptel--parse-schema backend gptel--schema)))
;; Merge request params with model and backend params.
(gptel--merge-plists
prompts-plist
@@ -313,6 +321,14 @@ Mutate state INFO with response metadata."
(gptel-backend-request-params gptel-backend)
(gptel--model-request-params gptel-model))))
+(cl-defmethod gptel--parse-schema ((_backend gptel-openai) schema)
+ (list :type "json_schema"
+ :json_schema
+ (list :name (md5 (format "%s" (random)))
+ :schema (gptel--preprocess-schema
+ (gptel--dispatch-schema-type schema))
+ :strict t)))
+
;; NOTE: No `gptel--parse-tools' method required for gptel-openai, since this is
;; handled by its defgeneric implementation
diff --git a/gptel-org.el b/gptel-org.el
index caa870b6..19bc62e9 100644
--- a/gptel-org.el
+++ b/gptel-org.el
@@ -171,8 +171,8 @@ adding elements to this list can significantly slow down
(defun gptel-org-set-topic (topic)
"Set a TOPIC and limit this conversation to the current heading.
-This limits the context sent to the LLM to the text between the
-current heading and the cursor position."
+This limits the context sent to the LLM to the text between the current
+heading (i.e. the heading with the topic set) and the cursor position."
(interactive
(list
(progn
@@ -248,6 +248,8 @@ depend on the value of `gptel-org-branching-context', which see."
(gptel-org--unescape-tool-results)
(gptel-org--strip-block-headers)
(when gptel-org-ignore-elements (gptel-org--strip-elements))
+ (setq org-complex-heading-regexp ;For org-element-context to run
+ (buffer-local-value 'org-complex-heading-regexp org-buf))
(current-buffer))))
;; Create prompt the usual way
(let ((org-buf (current-buffer))
@@ -256,6 +258,8 @@ depend on the value of `gptel-org-branching-context', which see."
(gptel-org--unescape-tool-results)
(gptel-org--strip-block-headers)
(when gptel-org-ignore-elements (gptel-org--strip-elements))
+ (setq org-complex-heading-regexp ;For org-element-context to run
+ (buffer-local-value 'org-complex-heading-regexp org-buf))
(current-buffer))))))
(defun gptel-org--strip-elements ()
diff --git a/gptel-rewrite.el b/gptel-rewrite.el
index 0ed0819c..240cd19f 100644
--- a/gptel-rewrite.el
+++ b/gptel-rewrite.el
@@ -28,9 +28,12 @@
(defvar eldoc-documentation-functions)
(defvar diff-entire-buffers)
+(defvar ediff-window-setup-function)
+(defvar ediff-split-window-function)
(declare-function diff-no-select "diff")
(declare-function rmc--add-key-description "rmc")
+(declare-function ediff-setup-windows-plain "ediff-wind")
;; * User options
@@ -310,6 +313,7 @@ BUF is the buffer to modify, defaults to the overlay buffer."
(when-let* ((overlay-buffer ov))
(let ((disp (overlay-get ov 'display))
(stored (overlay-get ov 'gptel--ediff)))
+ (overlay-put ov 'face (and restore 'gptel-rewrite-highlight-face))
(overlay-put ov 'display (and restore stored))
(overlay-put ov 'gptel--ediff (unless restore disp)))))))
(gptel--ediff-restore
@@ -320,7 +324,45 @@ BUF is the buffer to modify, defaults to the overlay buffer."
(remove-hook 'ediff-quit-hook gptel--ediff-restore))))
(funcall hideshow)
(add-hook 'ediff-quit-hook gptel--ediff-restore)
- (ediff-buffers ov-buf newbuf))))
+ (let ((ediff-window-setup-function #'ediff-setup-windows-plain)
+ (ediff-split-window-function #'split-window-horizontally))
+ (ediff-buffers ov-buf newbuf)))))
+
+(defun gptel--rewrite-merge-git (beg end new-str)
+ "Produce a merge conflict region between BEG and END.
+
+Merge the region with NEW-STR using git merge-file."
+ (let ((original-temp-file (make-temp-file "gptel-merge-"))
+ (empty-temp-file (make-temp-file "gptel-merge-")) ; use /dev/null? (windows?)
+ (new-temp-file (make-temp-file "gptel-merge-")))
+ (unwind-protect
+ (progn (write-region beg end original-temp-file)
+ (with-temp-file empty-temp-file (insert ""))
+ (with-temp-file new-temp-file (insert new-str))
+ (goto-char beg)
+ (delete-region beg end)
+ (call-process
+ "git" nil (list (current-buffer) nil) nil
+ "merge-file" "--no-diff3" "-L" "original" "-L" "Empty" "-L"
+ (gptel-backend-name gptel-backend) "-p"
+ original-temp-file empty-temp-file new-temp-file)
+ ;; Make merge marker active if required
+ (goto-char beg) (unless (bolp) (insert "\n")))
+ (delete-file original-temp-file)
+ (delete-file empty-temp-file)
+ (delete-file new-temp-file))))
+
+(defun gptel--rewrite-merge-simple (beg end new-str)
+ "Produce a merge conflict region between BEG and END.
+
+NEW-STR is the new string intended to replace the region."
+ (goto-char end) ;End first to preserve ordering
+ (unless (bolp) (insert "\n"))
+ (insert "=======\n" new-str "\n>>>>>>> "
+ (gptel-backend-name gptel-backend) "\n")
+ (goto-char beg)
+ (unless (bolp) (insert "\n"))
+ (insert-before-markers "<<<<<<< original\n"))
(defun gptel--rewrite-merge (&optional ovs)
"Insert pending LLM responses in OVS as merge conflicts."
@@ -332,15 +374,9 @@ BUF is the buffer to modify, defaults to the overlay buffer."
(dolist (ov (ensure-list ovs))
(save-excursion
(when-let* ((new-str (overlay-get ov 'gptel-rewrite)))
- ;; Insert merge
- (goto-char (overlay-start ov))
- (unless (bolp) (insert "\n"))
- (insert-before-markers "<<<<<<< original\n")
- (goto-char (overlay-end ov))
- (unless (bolp) (insert "\n"))
- (insert
- "=======\n" new-str
- "\n>>>>>>> " (gptel-backend-name gptel-backend) "\n")
+ (if (executable-find "git") ;Replace overlay content with merge result
+ (gptel--rewrite-merge-git (overlay-start ov) (overlay-end ov) new-str)
+ (gptel--rewrite-merge-simple (overlay-start ov) (overlay-end ov) new-str))
(setq changed t))))
(when changed (smerge-mode 1)))
(gptel--rewrite-reject ovs))))
diff --git a/gptel-transient.el b/gptel-transient.el
index 3d49bff1..5b0da5b8 100644
--- a/gptel-transient.el
+++ b/gptel-transient.el
@@ -793,7 +793,26 @@ Also format its value in the Transient menu."
(gptel--inspect-query
(gptel--suffix-send
(cons "I" (transient-args transient-current-command)))
- 'json)))]]
+ 'json)))]
+ ["Logging"
+ :if (lambda () (or gptel-log-level gptel-expert-commands))
+ ("-l" "Log level" "-l"
+ :class gptel-lisp-variable
+ :variable gptel-log-level
+ :set-value gptel--set-with-scope
+ :display-nil "Off"
+ :prompt "Log level: "
+ :reader
+ (lambda (prompt _ _)
+ "Manage gptel's logging."
+ (let ((state (completing-read
+ prompt '("off" "info" "debug") nil t)))
+ (message "Log level set to %s" state)
+ (if (string= state "off") nil (intern state)))))
+ ("L" "Inspect Log"
+ (lambda () (interactive)
+ (pop-to-buffer (get-buffer-create gptel--log-buffer-name)))
+ :format " %k %d")]]
[(gptel--suffix-send)]
(interactive)
(gptel--sanitize-model)
diff --git a/gptel.el b/gptel.el
index 4f70524e..24a1a086 100644
--- a/gptel.el
+++ b/gptel.el
@@ -35,7 +35,7 @@
;; gptel supports:
;;
;; - The services ChatGPT, Azure, Gemini, Anthropic AI, Together.ai, Perplexity,
-;; Anyscale, OpenRouter, Groq, PrivateGPT, DeepSeek, Cerebras, Github Models,
+;; AI/ML API, Anyscale, OpenRouter, Groq, PrivateGPT, DeepSeek, Cerebras, Github Models,
;; GitHub Copilot chat, AWS Bedrock, Novita AI, xAI, Sambanova, Mistral Le
;; Chat and Kagi (FastGPT & Summarizer).
;; - Local models via Ollama, Llama.cpp, Llamafiles or GPT4All
@@ -73,7 +73,7 @@
;; - For Gemini: define a gptel-backend with `gptel-make-gemini', which see.
;; - For Anthropic (Claude): define a gptel-backend with `gptel-make-anthropic',
;; which see.
-;; - For Together.ai, Anyscale, Groq, OpenRouter, DeepSeek, Cerebras or
+;; - For AI/ML API, Together.ai, Anyscale, Groq, OpenRouter, DeepSeek, Cerebras or
;; Github Models: define a gptel-backend with `gptel-make-openai', which see.
;; - For PrivateGPT: define a backend with `gptel-make-privategpt', which see.
;; - For Perplexity: define a backend with `gptel-make-perplexity', which see.
@@ -914,6 +914,11 @@ These parameters are combined with model-specific and backend-specific
incompatible with the active backend can break gptel. Do not use this
variable unless you know what you're doing!")
+(defconst gptel--ersatz-json-tool "response_json"
+ "Name of ersatz tool used to force JSON output.
+
+Some APIs, like Anthropic, use a tool to produce structured JSON output.")
+
;;; Utility functions
@@ -991,7 +996,7 @@ Later plists in the sequence take precedence over earlier ones."
(defun gptel--insert-file-string (path)
"Insert at point the contents of the file at PATH as context."
- (insert (format "In file `%s`:" (file-name-nondirectory path))
+ (insert (format "In file `%s`:" (abbreviate-file-name path))
"\n\n```\n")
(let ((pm (point-marker)))
(set-marker-insertion-type pm t)
@@ -1082,6 +1087,7 @@ Compatibility macro for Emacs 27.1."
"Copy gptel's local variables from BUF to a temp buffer and run BODY.
If positions START and END are provided, insert that part of BUF first."
+ (declare (indent 3))
`(gptel--with-buffer-copy-internal ,buf ,start ,end (lambda () ,@body)))
(defun gptel--with-buffer-copy-internal (buf start end body-thunk)
@@ -1092,7 +1098,7 @@ For BUF, START, END and BODY-THUNK see `gptel--with-buffer-copy'."
(with-current-buffer temp-buffer
(dolist (sym '( gptel-backend gptel--system-message gptel-model
gptel-mode gptel-track-response gptel-track-media
- gptel-use-tools gptel-tools gptel-use-curl
+ gptel-use-tools gptel-tools gptel-use-curl gptel--schema
gptel-use-context gptel--num-messages-to-send
gptel-stream gptel-include-reasoning gptel--request-params
gptel-temperature gptel-max-tokens gptel-cache))
@@ -1576,6 +1582,60 @@ file."
(declare-function gptel-context--wrap "gptel-context")
+
+;;; Structured output
+(defvar gptel--schema nil
+ "Response output schema for backends that support it.")
+
+(cl-defgeneric gptel--parse-schema (_backend _schema)
+ "Parse JSON schema in a backend-appropriate way.")
+
+(defun gptel--dispatch-schema-type (schema)
+ "Convert SCHEMA to a valid elisp representation."
+ (when (stringp schema)
+ (setq schema (gptel--json-read-string schema)))
+ ;; The OpenAI and Anthropic APIs don't allow arrays at the root of the schema.
+ ;; Work around this by wrapping it in an object with the field "items".
+ ;; TODO(schema): Find some way to strip this extra layer from the response.
+ (if (member (plist-get schema :type) '("array" array))
+ (list :type "object"
+ :properties (list :items schema)
+ :required ["items"]
+ :additionalProperties :json-false)
+ schema))
+
+(defun gptel--preprocess-schema (spec)
+ "Set additionalProperties for objects in SPEC destructively.
+
+Convert symbol :types to strings."
+ ;; NOTE: Do not use `sequencep' here, as that covers strings too and breaks
+ ;; things.
+ (when (or (listp spec) (vectorp spec))
+ (cond
+ ((vectorp spec)
+ (cl-loop for element across spec
+ for idx upfrom 0
+ do (aset spec idx (gptel--preprocess-schema element))))
+ ((keywordp (car spec))
+ (let ((tail spec))
+ (while tail
+ (when (eq (car tail) :type)
+ (when (symbolp (cadr tail)) ;Convert symbol :type to string
+ (setcar (cdr tail) (symbol-name (cadr tail))))
+ (when (equal (cadr tail) "object") ;Add additional object fields
+ (plist-put tail :additionalProperties :json-false)
+ (let ((props
+ (cl-loop for prop in (plist-get tail :properties) by #'cddr
+ collect (substring (symbol-name prop) 1))))
+ (plist-put tail :required (vconcat props)))))
+ (when (or (listp (cadr tail)) (vectorp (cadr tail)))
+ (gptel--preprocess-schema (cadr tail)))
+ (setq tail (cddr tail)))))
+ ((listp spec) (dolist (element spec)
+ (when (listp element)
+ (gptel--preprocess-schema element))))))
+ spec)
+
;;; Tool use
@@ -1867,7 +1927,7 @@ implementation, used by OpenAI-compatible APIs and Ollama."
(plist-get arg :name)))
(gptel-tool-args tool))))
:additionalProperties :json-false))
- (list :parameters :null)))))
+ (list :parameters (list :type "object" :properties nil))))))
(ensure-list tools))))
(cl-defgeneric gptel--parse-tool-results (backend results)
@@ -2242,7 +2302,12 @@ Run post-response hooks."
(cons 'tool-result result-alist) info)
(gptel--fsm-transition fsm)))))
(if (null tool-spec)
- (message "Unknown tool called by model: %s" name)
+ (if (equal name gptel--ersatz-json-tool) ;Could be a JSON response
+ ;; Handle structured JSON output supplied as tool call
+ (funcall (plist-get info :callback)
+ (gptel--json-encode (plist-get tool-call :args))
+ info)
+ (message "Unknown tool called by model: %s" name))
(setq arg-values
(mapcar
(lambda (arg)
@@ -2277,11 +2342,9 @@ Run post-response hooks."
(defun gptel--error-p (info) (plist-get info :error))
-(defun gptel--tool-use-p (info)
- (and (plist-get info :tools) (plist-get info :tool-use)))
+(defun gptel--tool-use-p (info) (plist-get info :tool-use))
-(defun gptel--tool-result-p (info)
- (and (plist-get info :tools) (plist-get info :tool-success)))
+(defun gptel--tool-result-p (info) (plist-get info :tool-success))
;; TODO(prompt-list): Document new prompt input format to `gptel-request'.
@@ -2292,7 +2355,7 @@ Run post-response hooks."
position context dry-run
(stream nil) (in-place nil)
(system gptel--system-message)
- transforms (fsm (gptel-make-fsm)))
+ schema transforms (fsm (gptel-make-fsm)))
"Request a response from the `gptel-backend' for PROMPT.
The request is asynchronous, this function returns immediately.
@@ -2442,6 +2505,13 @@ additional information (such as from a RAG engine).
and the state machine. It should run the callback after finishing its
transformation.
+If provided, SCHEMA forces the LLM to generate JSON output. Its value
+is a JSON schema, which can be provided as an elisp object, a nested
+plist structure. See the manual or the wiki for examples.
+
+Note: SCHEMA is presently experimental and subject to change, and not
+all providers support structured output.
+
See `gptel-prompt-transform-functions' for more.
FSM is the state machine driving the request. This can be used
@@ -2470,6 +2540,7 @@ be used to rerun or continue the request at a later time."
((markerp position) position)
((integerp position)
(set-marker (make-marker) position buffer))))
+ (gptel--schema schema)
(prompt-buffer
(cond ;prompt from buffer or explicitly supplied
((null prompt)
@@ -2702,6 +2773,8 @@ JSON query instead of the Lisp structure gptel uses."
(make-composed-keymap
(define-keymap
"C-c C-c" #'gptel--continue-query
+ "C-c C-w" (lambda () "Copy Curl command for query."
+ (interactive) (gptel--continue-query 'copy))
"C-c C-k" #'quit-window)
(current-local-map)))
(unless header-line-format
@@ -2710,17 +2783,24 @@ JSON query instead of the Lisp structure gptel uses."
(concat
"Edit request: \\[read-only-mode],"
" Send request: \\[gptel--continue-query],"
+ (format " Copy Curl: %s"
+ (propertize "C-c C-w" 'face 'help-key-binding))
" Quit: \\[quit-window]"))))
(display-buffer (current-buffer) gptel-display-buffer-action)))))
-(defun gptel--continue-query ()
+(declare-function gptel-curl--get-args "gptel-curl")
+
+(defun gptel--continue-query (&optional copy)
"Continue sending the gptel query displayed in this buffer.
The request is continued with the same parameters as originally
-specified."
- (interactive nil lisp-data-mode fundamental-mode)
+specified.
+
+With prefix arg COPY, copy the Curl command for the request to the
+kill ring instead."
+ (interactive "P" lisp-data-mode fundamental-mode)
(unless (equal (buffer-name) "*gptel-query*")
- (user-error "This command is meant for use in a gptel dry-run buffer."))
+ (user-error "This command is meant for use in a gptel dry-run buffer"))
(save-excursion
(goto-char (point-min))
(condition-case-unless-debug nil
@@ -2729,8 +2809,16 @@ specified."
(gptel--json-read))))
(cl-assert (cl-typep gptel--fsm-last 'gptel-fsm))
(plist-put (gptel-fsm-info gptel--fsm-last) :data data)
- (gptel--fsm-transition gptel--fsm-last) ;INIT -> WAIT
- (quit-window))
+ (if copy ;Copy Curl command instead of sending request
+ (let ((args (and (require 'gptel-curl)
+ (gptel-curl--get-args (gptel-fsm-info gptel--fsm-last)
+ (md5 (format "%s" (random)))))))
+ (kill-new
+ (mapconcat #'shell-quote-argument
+ (cons (gptel--curl-path) args) " \\\n"))
+ (message "Curl command for request copied to kill-ring"))
+ (gptel--fsm-transition gptel--fsm-last) ;INIT -> WAIT
+ (quit-window)))
(error
(user-error "Can not resume request: could not read data from buffer!")))))
@@ -3670,7 +3758,7 @@ PRESET is a spec (plist) of keys and values."
(while index
(setq key (pop index) val (pop index))
(pcase key
- (:description)
+ ((or :description :pre :post))
(:parents
(mapc (lambda (parent-preset)
(nconc syms (gptel--preset-syms
diff --git a/test b/test
index 3e644187..f33daff4 160000
--- a/test
+++ b/test
@@ -1 +1 @@
-Subproject commit 3e6441877dff02e71fb3538120094d994c5bd7df
+Subproject commit f33daff49c5ebbf991473906e2c83f0a867e0d8d