Add :sse_middleware option to allow custom SSE middleware for streaming

CHANGELOG.md

@@ -7,6 +7,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 ## [0.20.0] - 2026-06-29
 
 ### Added
+- Added `:sse_middleware` option for streaming responses, allowing consumers to inject a custom SSE middleware instead of the default `Tesla.Middleware.SSE`.
 - Added `LlmComposer.Agent` — an automatic tool-calling loop that runs the full ask → execute → repeat cycle until the model returns a final answer. Supports sequential and parallel tool execution, configurable iteration limit, and per-tool error recovery.
 - Added streaming support to `LlmComposer.Agent` (`stream_response: true`). Returns a lazy stream of `StreamChunk`s with `:text_delta`, `:tool_call` (one per executed tool, result included), and a terminal `:done` chunk carrying cumulative usage and the full `Agent.Result`. Supported providers: `:open_ai`, `:open_router`, `:open_ai_responses`, `:google`, `:bedrock`, `:ollama`.
 - Added `:tool_call` chunk type to `LlmComposer.StreamChunk` — emitted by the streaming agent after each tool execution, making the stream self-contained without needing a telemetry handler.
@@ -297,10 +298,10 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 - **Include Bedrock Support**: Included Bedrock support as provider only with `completion()`s support.
 
 ## [0.3.5] - 2024-12-18
-- **Fix default api_key setting value**: The default value was en empty string. Now it is nil to be evaluated as false when getting the key from the map later. 
+- **Fix default api_key setting value**: The default value was en empty string. Now it is nil to be evaluated as false when getting the key from the map later.
 
 ## [0.3.4] - 2024-12-17
-- **OpenAI API Keys**: Use an API key passed as a parameter when calling chat_completion — overriding the global API key defined in the config. The param is sent inside the settings. 
+- **OpenAI API Keys**: Use an API key passed as a parameter when calling chat_completion — overriding the global API key defined in the config. The param is sent inside the settings.
 
 ## [0.3.3] - 2024-12-04
 - **Timeouts**: Configurable OpenAI's timeout. Default set to 50 seconds.

lib/llm_composer/http_client.ex

@@ -47,7 +47,7 @@ defmodule LlmComposer.HttpClient do
 
     cond do
       stream ->
-        resp ++ [{Tesla.Middleware.SSE, only: :data}]
+        resp ++ Keyword.get(opts, :sse_middleware, [{Tesla.Middleware.SSE, only: :data}])
 
       retries_disabled?(opts) ->
         resp ++

lib/llm_composer/providers_runner.ex

@@ -115,8 +115,12 @@ defmodule LlmComposer.ProvidersRunner do
     |> Keyword.put_new(:stream_response, settings.stream_response)
     |> Keyword.put_new(:track_costs, settings.track_costs)
     |> Keyword.put_new(:api_key, settings.api_key)
+    |> maybe_put_sse_middleware(settings)
   end
 
+  defp maybe_put_sse_middleware(opts, %{sse_middleware: nil}), do: opts
+  defp maybe_put_sse_middleware(opts, %{sse_middleware: middleware}), do: Keyword.put_new(opts, :sse_middleware, middleware)
+
   defp get_provider_router do
     :llm_composer
     |> Application.get_env(:provider_router, [])

lib/llm_composer/settings.ex

@@ -9,6 +9,7 @@ defmodule LlmComposer.Settings do
             provider: nil,
             provider_opts: nil,
             providers: nil,
+            sse_middleware: nil,
             stream_response: false,
             system_prompt: nil,
             track_costs: false,
@@ -21,6 +22,7 @@ defmodule LlmComposer.Settings do
   - `:provider` — provider module (e.g. `LlmComposer.Providers.OpenAI`). Used with `:provider_opts`.
   - `:provider_opts` — keyword list of options forwarded to the single provider.
   - `:providers` — list of `{module, opts}` tuples for multi-provider routing via `ProvidersRunner`.
+  - `:sse_middleware` — custom Tesla SSE middleware list to use instead of the default `Tesla.Middleware.SSE` when `stream_response: true`. Useful when the provider sends incomplete SSE chunks that would crash the default middleware.
   - `:stream_response` — when `true`, the provider returns a streaming enumerable.
   - `:system_prompt` — system prompt string sent to the model.
   - `:track_costs` — when `true`, token usage and cost are computed and attached to the response.
@@ -31,6 +33,7 @@ defmodule LlmComposer.Settings do
           provider: module() | nil,
           provider_opts: keyword() | nil,
           providers: [{module(), keyword()}] | nil,
+          sse_middleware: [{module(), keyword()}] | nil,
           stream_response: boolean(),
           system_prompt: String.t() | nil,
           track_costs: boolean(),