feat: adapter and debug window improvement

Author: olimorris

codecompanion-workspace.json

@@ -81,6 +81,30 @@
           "path": "tests/helpers.lua"
         }
       ]
+    },
+    {
+      "name": "Adapters",
+      "system_prompt": "In the CodeCompanion plugin, adapters are used to connect to LLMs. The adapters contain various options for the LLM's endpoint alongside a defined schema for properties such as the model, temperature, top k, top p etc. The adapters also contain various handler functions which define how messages which are sent to the LLM should be formatted alongside how output from the LLM should be received and displayed in the chat buffer. The adapters are defined in the `adapters` directory.",
+      "opts": {
+        "remove_config_system_prompt": true
+      },
+      "vars": {
+        "base_dir": "lua/codecompanion"
+      },
+      "files": [
+        {
+          "description": "Each LLM has their own adapter. This allows for LLM settings to be generated from the schema table in an adapter before they're sent to the LLM via the http file. ",
+          "path": "${base_dir}/adapters/init.lua"
+        },
+        {
+          "description": "Adapters are then passed to the http client which sends requests to LLMs via Curl:",
+          "path": "${base_dir}/http.lua"
+        },
+        {
+          "description": "Adapters must follow a schema. The validation and how schema values are extracted from the table schema is defined in:",
+          "path": "${base_dir}/schema.lua"
+        }
+      ]
     }
   ]
 }

doc/extending/adapters.md

@@ -18,6 +18,8 @@ Let's take a look at the interface of an adapter as per the `adapter.lua` file:
 ---@field env_replaced? table Replacement of environment variables with their actual values
 ---@field headers table The headers to pass to the request
 ---@field parameters table The parameters to pass to the request
+---@field body table Additional body parameters to pass to the request
+---@field chat_prompt string The system chat prompt to send to the LLM
 ---@field raw? table Any additional curl arguments to pass to the request
 ---@field opts? table Additional options for the adapter
 ---@field handlers table Functions which link the output from the request to CodeCompanion
@@ -448,4 +450,5 @@ temperature = {
 },
 ```
 
-You'll see we've specified a function call for the `condition` key. We're simply checking that the model name doesn't being with `o1` as these models don't accept temperature as a parameter. You'll also see we've specified a function call for the `validate` key. We're simply checking that the value of the temperature is between 0 and 2
+You'll see we've specified a function call for the `condition` key. We're simply checking that the model name doesn't being with `o1` as these models don't accept temperature as a parameter. You'll also see we've specified a function call for the `validate` key. We're simply checking that the value of the temperature is between 0 and 2.
+

lua/codecompanion/adapters/anthropic.lua

@@ -18,6 +18,11 @@ return {
     text = true,
     vision = true,
   },
+  opts = {
+    stream = true,
+    cache_breakpoints = 4, -- Cache up to this many messages
+    cache_over = 300, -- Cache any message which has this many tokens or more
+  },
   url = "https://api.anthropic.com/v1/messages",
   env = {
     api_key = "ANTHROPIC_API_KEY",
@@ -28,15 +33,17 @@ return {
     ["anthropic-version"] = "2023-06-01",
     ["anthropic-beta"] = "prompt-caching-2024-07-31",
   },
-  parameters = {
-    stream = true,
-  },
-  opts = {
-    stream = true, -- NOTE: Currently, CodeCompanion ONLY supports streaming with this adapter
-    cache_breakpoints = 4, -- Cache up to this many messages
-    cache_over = 300, -- Cache any message which has this many tokens or more
-  },
   handlers = {
+    ---@param self CodeCompanion.Adapter
+    ---@return boolean
+    setup = function(self)
+      if self.opts and self.opts.stream then
+        self.parameters.stream = true
+      end
+
+      return true
+    end,
+
     ---Set the parameters
     ---@param self CodeCompanion.Adapter
     ---@param params table
@@ -116,11 +123,15 @@ return {
 
     ---Returns the number of tokens generated from the LLM
     ---@param self CodeCompanion.Adapter
-    ---@param data string The data from the LLM
+    ---@param data table The data from the LLM
     ---@return number|nil
     tokens = function(self, data)
       if data then
-        data = data:sub(6)
+        if self.opts.stream then
+          data = utils.clean_streamed_data(data)
+        else
+          data = data.body
+        end
         local ok, json = pcall(vim.json.decode, data)
 
         if ok then
@@ -129,29 +140,36 @@ return {
               + (json.message.usage.cache_creation_input_tokens or 0)
 
             output_tokens = json.message.usage.output_tokens or 0
-          end
-          if json.type == "message_delta" then
+          elseif json.type == "message_delta" then
             return (input_tokens + output_tokens + json.usage.output_tokens)
+          elseif json.type == "message" then
+            return (json.usage.input_tokens + json.usage.output_tokens)
           end
         end
       end
     end,
 
     ---Output the data from the API ready for insertion into the chat buffer
     ---@param self CodeCompanion.Adapter
-    ---@param data string The streamed JSON data from the API, also formatted by the format_data handler
+    ---@param data table The streamed JSON data from the API, also formatted by the format_data handler
     ---@return table|nil
     chat_output = function(self, data)
       local output = {}
 
-      -- Skip the event messages
-      if type(data) == "string" and string.sub(data, 1, 6) == "event:" then
-        return
+      if self.opts.stream then
+        if type(data) == "string" and string.sub(data, 1, 6) == "event:" then
+          return
+        end
       end
 
       if data and data ~= "" then
-        local data_mod = utils.clean_streamed_data(data)
-        local ok, json = pcall(vim.json.decode, data_mod, { luanil = { object = true } })
+        if self.opts.stream then
+          data = utils.clean_streamed_data(data)
+        else
+          data = data.body
+        end
+
+        local ok, json = pcall(vim.json.decode, data, { luanil = { object = true } })
 
         if ok then
           if json.type == "message_start" then
@@ -160,6 +178,9 @@ return {
           elseif json.type == "content_block_delta" then
             output.role = nil
             output.content = json.delta.text
+          elseif json.type == "message" then
+            output.role = json.role
+            output.content = json.content[1].text
           end
 
           return {
@@ -176,17 +197,25 @@ return {
     ---@param context table Useful context about the buffer to inline to
     ---@return table|nil
     inline_output = function(self, data, context)
-      if type(data) == "string" and string.sub(data, 1, 6) == "event:" then
-        return
+      if self.opts.stream then
+        if type(data) == "string" and string.sub(data, 1, 6) == "event:" then
+          return
+        end
       end
 
       if data and data ~= "" then
-        data = data:sub(6)
+        if self.opts.stream then
+          data = utils.clean_streamed_data(data)
+        else
+          data = data.body
+        end
         local ok, json = pcall(vim.json.decode, data, { luanil = { object = true } })
 
         if ok then
           if json.type == "content_block_delta" then
             return json.delta.text
+          elseif json.type == "message" then
+            return json.content[1].text
           end
         end
       end
@@ -257,7 +286,7 @@ return {
       default = nil,
       desc = "Only sample from the top K options for each subsequent token. Use top_k to remove long tail low probability responses",
       validate = function(n)
-        return n >= 0 and n <= 500, "Must be between 0 and 500"
+        return n >= 0, "Must be greater than 0"
       end,
     },
     stop_sequences = {

lua/codecompanion/adapters/copilot.lua

@@ -254,7 +254,7 @@ return {
       order = 4,
       mapping = "parameters",
       type = "integer",
-      default = 4096,
+      default = 15000,
       desc = "The maximum number of tokens to generate in the chat completion. The total length of input tokens and generated tokens is limited by the model's context length.",
     },
     top_p = {

lua/codecompanion/adapters/deepseek.lua

@@ -1,4 +1,3 @@
-local log = require("codecompanion.utils.log")
 local openai = require("codecompanion.adapters.openai")
 local utils = require("codecompanion.utils.adapters")
 
@@ -90,15 +89,12 @@ return {
             if delta.role then
               output.role = delta.role
             end
-
             if self.opts.can_reason and delta.reasoning_content then
               output.reasoning = delta.reasoning_content
             end
-
             if delta.content then
               output.content = (output.content or "") .. delta.content
             end
-
             return {
               status = "success",
               output = output,

lua/codecompanion/adapters/gemini.lua

@@ -2,6 +2,7 @@
 ---https://github.com/google-gemini/cookbook/blob/main/quickstarts/rest/Streaming_REST.ipynb
 
 local log = require("codecompanion.utils.log")
+local utils = require("codecompanion.utils.adapters")
 
 ---@class Gemini.Adapter: CodeCompanion.Adapter
 return {
@@ -12,17 +13,25 @@ return {
     user = "user",
   },
   opts = {
-    stream = true, -- NOTE: Currently, CodeCompanion ONLY supports streaming with this adapter
+    stream = true,
   },
   features = {
     tokens = true,
     text = true,
     vision = true,
   },
-  url = "https://generativelanguage.googleapis.com/v1beta/models/${model}:streamGenerateContent?alt=sse&key=${api_key}",
+  url = "https://generativelanguage.googleapis.com/v1beta/models/${model}${stream}key=${api_key}",
   env = {
     api_key = "GEMINI_API_KEY",
     model = "schema.model.default",
+    stream = function(self)
+      local stream = ":generateContent?"
+      if self.opts.stream then
+        -- NOTE: With sse each stream chunk is a GenerateContentResponse object with a portion of the output text in candidates[0].content.parts[0].text
+        stream = ":streamGenerateContent?alt=sse&"
+      end
+      return stream
+    end,
   },
   headers = {
     ["Content-Type"] = "application/json",
@@ -93,7 +102,7 @@ return {
     ---@return number|nil
     tokens = function(self, data)
       if data and data ~= "" then
-        data = data:sub(6)
+        data = utils.clean_streamed_data(data)
         local ok, json = pcall(vim.json.decode, data, { luanil = { object = true } })
 
         if ok then
@@ -110,7 +119,7 @@ return {
       local output = {}
 
       if data and data ~= "" then
-        data = data:sub(6)
+        data = utils.clean_streamed_data(data)
         local ok, json = pcall(vim.json.decode, data, { luanil = { object = true } })
 
         if ok and json.candidates[1].content then
@@ -132,7 +141,7 @@ return {
     ---@return table|nil
     inline_output = function(self, data, context)
       if data and data ~= "" then
-        data = data:sub(6)
+        data = utils.clean_streamed_data(data)
         local ok, json = pcall(vim.json.decode, data, { luanil = { object = true } })
 
         if not ok then
@@ -176,5 +185,65 @@ return {
         "gemini-1.0-pro",
       },
     },
+    maxOutputTokens = {
+      order = 2,
+      mapping = "body.generationConfig",
+      type = "integer",
+      optional = true,
+      default = nil,
+      desc = "The maximum number of tokens to include in a response candidate. Note: The default value varies by model",
+      validate = function(n)
+        return n > 0, "Must be greater than 0"
+      end,
+    },
+    temperature = {
+      order = 3,
+      mapping = "body.generationConfig",
+      type = "number",
+      optional = true,
+      default = nil,
+      desc = "Controls the randomness of the output.",
+      validate = function(n)
+        return n >= 0 and n <= 2, "Must be between 0 and 2"
+      end,
+    },
+    topP = {
+      order = 4,
+      mapping = "body.generationConfig",
+      type = "integer",
+      optional = true,
+      default = nil,
+      desc = "The maximum cumulative probability of tokens to consider when sampling. The model uses combined Top-k and Top-p (nucleus) sampling. Tokens are sorted based on their assigned probabilities so that only the most likely tokens are considered. Top-k sampling directly limits the maximum number of tokens to consider, while Nucleus sampling limits the number of tokens based on the cumulative probability.",
+      validate = function(n)
+        return n > 0, "Must be greater than 0"
+      end,
+    },
+    topK = {
+      order = 5,
+      mapping = "body.generationConfig",
+      type = "integer",
+      optional = true,
+      default = nil,
+      desc = "The maximum number of tokens to consider when sampling",
+      validate = function(n)
+        return n > 0, "Must be greater than 0"
+      end,
+    },
+    presencePenalty = {
+      order = 6,
+      mapping = "body.generationConfig",
+      type = "number",
+      optional = true,
+      default = nil,
+      desc = "Presence penalty applied to the next token's logprobs if the token has already been seen in the response",
+    },
+    frequencyPenalty = {
+      order = 7,
+      mapping = "body.generationConfig",
+      type = "number",
+      optional = true,
+      default = nil,
+      desc = "Frequency penalty applied to the next token's logprobs, multiplied by the number of times each token has been seen in the response so far.",
+    },
   },
 }