Switching to OpenRouter.

Sixzero · Sixzero · commit 442dc9a85414 · 2025-11-25T09:47:06.000+01:00
diff --git a/Project.toml b/Project.toml
@@ -1,7 +1,7 @@
 name = "EasyContext"
 uuid = "5067fbba-b9ec-41e5-aa31-a8c1ac509a30"
-authors = ["SixZero <havliktomi@hotmail.com>"]
 version = "0.6.1"
+authors = ["SixZero <havliktomi@hotmail.com>"]
 
 [deps]
 Arrow = "69666777-d1a9-59fb-9406-91d4454c9d45"
@@ -22,6 +22,7 @@ LLMRateLimiters = "ed1ca689-44f5-49fd-ae5c-d523230a773a"
 LibGit2 = "76f85450-5226-5b5a-8eaa-529ad045b433"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 Markdown = "d6f4376e-aef5-505a-96c1-9c027394607a"
+OpenRouter = "b1229eab-bc7b-489c-92b5-04baf87d8d0b"
 Parameters = "d96e819e-fc66-5662-9728-84c9c7592b0a"
 Pkg = "44cfe95a-1eb2-52ea-b672-e2afdf69b78f"
 PrecompileTools = "aea7be01-6a6a-4083-8856-8a6e6704d82a"
@@ -34,7 +35,6 @@ Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 SHA = "ea8e919c-243c-51af-8825-aaa63cd721ce"
 Snowball = "fb8f903a-0164-4e73-9ffe-431110250c3b"
 SparseArrays = "2f01184e-e22b-5df5-ae63-d93ebab69eaf"
-StreamCallbacksExt = "318dcf79-f1de-4874-b31b-6e12fc76f75a"
 ULID = "28ea79d9-6354-52a7-8af1-38e01db141d4"
 UUIDs = "cf7118a7-6976-5b1a-9a39-7adc72f591a4"
 
@@ -66,7 +66,6 @@ SparseArrays = "≥1.9"
 LLMRateLimiters = "≥1.1.0"
 Crayons = "≥4.1"
 ProgressMeter = "≥1.10"
-StreamCallbacksExt = "≥0.4.5"
 Dates = "≥1.11"
 ULID = "≥0.3"
 JLD2 = "≥0.5"
diff --git a/archived/automation/condition_llm.jl b/archived/automation/condition_llm.jl
@@ -1,4 +1,3 @@
-using PromptingTools
 
 """
     condition_llm(condition::AbstractString; model::AbstractString="gpt4om", verbose::Bool=false)
diff --git a/archived/automation/selector_llm.jl b/archived/automation/selector_llm.jl
@@ -1,4 +1,3 @@
-using PromptingTools
 
 export selector_llm
 
diff --git a/examples/fluid_agent_usage.jl b/examples/fluid_agent_usage.jl
@@ -1,5 +1,4 @@
 using EasyContext
-using PromptingTools
 
 """
 This example demonstrates various ways to use FluidAgent for automation tasks.
diff --git a/examples/julia_search_usage.jl b/examples/julia_search_usage.jl
@@ -1,6 +1,5 @@
 using EasyContext
 using EasyContext: push_message!, BM25Embedder, TopK, julia_ctx_2_string, get_source
-using PromptingTools
 
 """
 This example demonstrates how to use JuliaSearchTool and JuliaCTX for searching Julia code.
diff --git a/src/EasyContext.jl b/src/EasyContext.jl
@@ -7,8 +7,8 @@ using LLMRateLimiters: TokenEstimationMethod, airatelimited, RateLimiterRPM, Rat
 using Dates
 using UUIDs
 
-using PromptingTools
-using PromptingTools: UserMessage, AIMessage, SystemMessage
+using OpenRouter
+using OpenRouter: UserMessage, AIMessage, SystemMessage
 
 export search
 export EFFICIENT_PIPELINE, HIGH_ACCURACY_PIPELINE
diff --git a/src/Rephrase.v1.jl b/src/Rephrase.v1.jl
@@ -1,7 +1,6 @@
 
 using RAGTools: AbstractRephraser
 import RAGTools: rephrase
-using PromptingTools
 
 Base.@kwdef struct JuliacodeRephraser <: AbstractRephraser
   template::Symbol=:RAGRephraserByKeywords
@@ -16,7 +15,7 @@ function rephrase(rephraser::JuliacodeRephraser, question::AbstractString;
   placeholders = only(aitemplates(template)).variables # only one template should be found
   @assert (:query in placeholders) "Provided RAG Template $(template) is not suitable. It must have a placeholder: `query`."
 
-  msg = aigenerate(template; query = question, verbose, model, kwargs...)
+  msg = aigen(template, model; query = question, verbose, kwargs...)
   Threads.atomic_add!(cost_tracker, msg.cost)
   useful_doc_ideas, possible_files_to_edit = parse_and_evaluate_script(msg.content)
   @show length.((useful_doc_ideas, possible_files_to_edit))
diff --git a/src/agents/FluidAgent.jl b/src/agents/FluidAgent.jl
@@ -1,7 +1,6 @@
 using UUIDs
-using PromptingTools
-using PromptingTools: aigenerate
-using StreamCallbacksExt: needs_tool_execution
+import PromptingTools: aigenerate
+using OpenRouter: needs_tool_execution
 
 export FluidAgent, execute_tools, work, create_FluidAgent
 
diff --git a/src/chunkers/FullFileChunker.jl b/src/chunkers/FullFileChunker.jl
@@ -1,4 +1,4 @@
-using PromptingTools: recursive_splitter
+import PromptingTools: recursive_splitter
 using RAGTools
 using RAGTools: AbstractChunker
 
diff --git a/src/chunkers/NewlineChunker.jl b/src/chunkers/NewlineChunker.jl
@@ -1,5 +1,5 @@
 using FilePathsBase
-using PromptingTools: recursive_splitter
+import PromptingTools: recursive_splitter
 using RAGTools
 using RAGTools: AbstractChunker
 using PyCall
diff --git a/src/embedders/Embedders.jl b/src/embedders/Embedders.jl
@@ -1,4 +1,4 @@
-using PromptingTools
+import PromptingTools
 using LinearAlgebra
 using RAGTools
 using JLD2, Snowball, Pkg
diff --git a/src/embedders/api/CohereEmbedder.jl b/src/embedders/api/CohereEmbedder.jl
@@ -1,6 +1,5 @@
 using HTTP
 using JSON3
-using PromptingTools
 using RAGTools: get_embeddings
 using ProgressMeter
 using LLMRateLimiters
diff --git a/src/embedders/api/GoogleGeckoEmbedder.jl b/src/embedders/api/GoogleGeckoEmbedder.jl
@@ -1,6 +1,5 @@
 using HTTP
 using JSON3
-using PromptingTools
 using RAGTools: get_embeddings
 using ProgressMeter
 using LLMRateLimiters
diff --git a/src/embedders/api/JinaEmbedder.jl b/src/embedders/api/JinaEmbedder.jl
@@ -1,7 +1,6 @@
 
 using HTTP
 using JSON3
-using PromptingTools
 using RAGTools: get_embeddings
 using ProgressMeter
 
diff --git a/src/embedders/api/OpenAIBatchEmbedder.jl b/src/embedders/api/OpenAIBatchEmbedder.jl
@@ -1,6 +1,6 @@
 
 using RAGTools: BatchEmbedder
-using PromptingTools: MODEL_EMBEDDING
+import PromptingTools: MODEL_EMBEDDING
 using SparseArrays
 using LinearAlgebra: normalize
 
diff --git a/src/embedders/api/OpenAIBatchEmbedderWithRetry.jl b/src/embedders/api/OpenAIBatchEmbedderWithRetry.jl
@@ -1,9 +1,7 @@
 using RAGTools: BatchEmbedder, _normalize
-using PromptingTools: MODEL_EMBEDDING
+import PromptingTools: MODEL_EMBEDDING
 using SparseArrays
 using Base.Threads
-using PromptingTools
-const PT = PromptingTools
 
 @kwdef struct OpenAIBatchEmbedderWithRetry <: AbstractEasyEmbedder
     embedder::BatchEmbedder=BatchEmbedder()
diff --git a/src/embedders/api/VoyageEmbedder.jl b/src/embedders/api/VoyageEmbedder.jl
@@ -1,6 +1,5 @@
 using HTTP
 using JSON3
-using PromptingTools
 using RAGTools: get_embeddings
 using ProgressMeter
 using LLMRateLimiters
diff --git a/src/input/test.jl b/src/input/test.jl
@@ -3,7 +3,7 @@ init_assemblyai();
 println("Starting audio stream. Speak into your microphone. Press Ctrl+C to stop.");
 stream_audio(text -> println("Transcribed: ", text))
 #%%
-using PromptingTools: CerebrasOpenAISchema, aigenerate
+import PromptingTools: CerebrasOpenAISchema, aigenerate
 aigenerate(CerebrasOpenAISchema(), "Tell me a joke."; model="gpt-oss-120b") # works
 
 using StreamCallbacks
diff --git a/src/precompile_scripts.jl b/src/precompile_scripts.jl
@@ -1,6 +1,7 @@
 using EasyContext
-using PrecompileTools
+using OpenRouter
 using DataStructures: OrderedDict
+using PrecompileTools
 
 @setup_workload begin
     # Dummy values for testing
@@ -16,7 +17,7 @@ using DataStructures: OrderedDict
     ]
 
     # Mock AI function that doesn't make API calls
-    mock_ai_fn(prompt; kwargs...) = PromptingTools.AIMessage("1,2")
+    mock_ai_fn(prompt; kwargs...) = OpenRouter.AIMessage("1,2")
 
     @time "Precompilation EasyContext.jl" @compile_workload begin
         # Initialize workspace context
diff --git a/src/prompts/prompt_instant_apply.jl b/src/prompts/prompt_instant_apply.jl
@@ -1,4 +1,4 @@
-using PromptingTools: create_template
+import PromptingTools: create_template
 
 
 function get_merge_prompt_v1(original_content, changes_content)
diff --git a/src/protocol/Conversation.jl b/src/protocol/Conversation.jl
@@ -1,5 +1,4 @@
-using StreamCallbacksExt
-using StreamCallbacksExt: RunInfo
+using OpenRouter: RunInfo
 
 @kwdef mutable struct Conversation{M <: MSG} <: CONV
     system_message::M
@@ -64,12 +63,12 @@ function to_dict_nosys_detailed(conv::CONV)
     [to_dict(message) for message in conv.messages]
 end
 # just unwrap it.
-update_last_user_message_meta(conv::CONV, callback::StreamCallbackChannelWrapper) = update_last_user_message_meta(conv, callback.callback)
-function update_last_user_message_meta(conv::CONV, callback::Union{StreamCallbackWithHooks, StreamCallbackWithTokencounts})
-    tokens::TokenCounts, run_info::RunInfo, flavor, model = callback.total_tokens, callback.run_info, callback.flavor, callback.model
-    elapsed = run_info.last_message_time - run_info.creation_time
-    update_last_user_message_meta(conv, tokens.input, tokens.output, tokens.cache_write, tokens.cache_read, StreamCallbacksExt.get_cost(flavor, model, tokens), elapsed)
-end
+# update_last_user_message_meta(conv::CONV, callback::StreamCallbackChannelWrapper) = update_last_user_message_meta(conv, callback.callback)
+# function update_last_user_message_meta(conv::CONV, callback::Union{StreamCallbackWithHooks, StreamCallbackWithTokencounts})
+#     tokens::TokenCounts, run_info::RunInfo, flavor, model = callback.total_tokens, callback.run_info, callback.flavor, callback.model
+#     elapsed = run_info.last_message_time - run_info.creation_time
+#     update_last_user_message_meta(conv, tokens.input, tokens.output, tokens.cache_write, tokens.cache_read, OpenRouter.get_cost(flavor, model, tokens), elapsed)
+# end
 update_last_user_message_meta(conv::CONV, itok::Int, otok::Int, cached::Int, cache_read::Int, price, elapsed::Number) = update_message(conv.messages[end], itok, otok, cached, cache_read, price, elapsed)
 
 last_msg(conv::CONV) = conv.messages[end].content
diff --git a/src/protocol/Session.jl b/src/protocol/Session.jl
@@ -1,4 +1,6 @@
+using OpenRouter: AbstractMessage, SystemMessage, UserMessage, AIMessage
 export Session, initSession
+
 include("SessionImageSupport.jl")
 
 @kwdef mutable struct Session{M <: MSG} <: CONV
@@ -43,8 +45,9 @@ conversaion_file(path,conv::Session) = joinpath(conversaion_path(path, conv), "c
 
 
 function to_PT_messages(session::Session, sys_msg::String, imagepaths_in_messages_supported::Bool=false)
-    messages = Vector{PT.AbstractChatMessage}(undef, length(session.messages) + 1)
-    messages[1] = SystemMessage(sys_msg)
+    
+    messages = Vector{AbstractMessage}(undef, length(session.messages) + 1)
+    messages[1] = SystemMessage(content=sys_msg)
     
     for (i, msg) in enumerate(session.messages)
         messages[i + 1] = if msg.role == :user
@@ -55,23 +58,59 @@ function to_PT_messages(session::Session, sys_msg::String, imagepaths_in_message
             base64_images = filter(p -> startswith(p.first, "base64img_") || startswith(p.second, "data:image"), collect(msg.context))
             base64_urls = isempty(base64_images) ? String[] : [p.second for p in base64_images]
             
+            # Combine content with context
+            full_content = context_combiner!(msg.content, msg.context, false)
+            
+            # Create OpenRouter UserMessage with image data
             if !isempty(image_paths) || !isempty(base64_urls)
-                # Use the existing constructor which handles both paths and base64
-                new_content = context_combiner!(msg.content, msg.context, false)
-                PT.UserMessageWithImages(new_content; 
-                    image_path = isempty(image_paths) ? nothing : image_paths,
-                    image_url = isempty(base64_urls) ? nothing : base64_urls)
+                # Convert file paths to data URLs if needed
+                all_image_data = String[]
+                for path in image_paths
+                    if isfile(path)
+                        # Convert file to data URL (you may need to implement this helper)
+                        push!(all_image_data, file_to_data_url(path))
+                    end
+                end
+                append!(all_image_data, base64_urls)
+                
+                UserMessage(content=full_content, image_data=all_image_data)
             else
-                full_content = context_combiner!(msg.content, msg.context)
-                UserMessage(full_content)
+                UserMessage(content=full_content)
             end
         elseif msg.role == :assistant 
             full_content = context_combiner!(msg.content, msg.context)
-            AIMessage(full_content)
+            AIMessage(content=full_content)
         else
             full_content = context_combiner!(msg.content, msg.context)
-            UserMessage(full_content)
+            UserMessage(content=full_content)
         end
     end
     return messages
+end
+
+# Helper function to convert file path to data URL (add this if not exists)
+function file_to_data_url(filepath::String)
+    if !isfile(filepath)
+        return ""
+    end
+    
+    # Determine MIME type based on extension
+    ext = lowercase(splitext(filepath)[2])
+    mime_type = if ext == ".jpg" || ext == ".jpeg"
+        "image/jpeg"
+    elseif ext == ".png"
+        "image/png"
+    elseif ext == ".gif"
+        "image/gif"
+    elseif ext == ".webp"
+        "image/webp"
+    else
+        "image/jpeg"  # default
+    end
+    
+    # Read file and encode as base64
+    data = read(filepath)
+    b64_data = base64encode(data)
+    
+    return "data:$mime_type;base64,$b64_data"
 end
diff --git a/src/rag/AdvancedRAG.jl b/src/rag/AdvancedRAG.jl
@@ -72,14 +72,14 @@ Uses Cohere embedder and BM25 for retrieval with top_k=50, and ReduceGPTReranker
 - `cache_prefix::String="workspace"`: Prefix for the embedder cache
 - `verbose::Int=0`: Verbosity level (0=quiet, 1=normal, 2=detailed). If >1, returns timing-enabled pipeline.
 """
-function EFFICIENT_PIPELINE(; top_n=10, top_k=50, rerank_prompt=create_rankgpt_prompt_v2, model::Union{AbstractString,Vector{String},ModelConfig}=["gemf", "grokfast"], cache_prefix="prefix", verbose=1)
+function EFFICIENT_PIPELINE(; top_n=10, top_k=20, rerank_prompt=create_rankgpt_prompt_v2, model::Union{AbstractString,Vector{String},ModelConfig}=["gemf", "grokfast"], cache_prefix="prefix", verbose=1)
     # embedder = create_openai_embedder(cache_prefix=cache_prefix)
     embedder_verbose = verbose > 0
     timing_enabled = verbose > 1
     embedder = EasyContext.create_cohere_embedder(model="embed-v4.0", cache_prefix=cache_prefix, verbose=embedder_verbose)
     bm25 = BM25Embedder()
     topK = TopK([embedder, bm25]; top_k)
-    reranker = ReduceGPTReranker(batch_size=30; top_n, model, rerank_prompt, verbose)
+    reranker = ReduceGPTReranker(batch_size=12; top_n, model, rerank_prompt, verbose)
     
     return timing_enabled ? TwoLayerRAGWithTimings(; topK, reranker) : TwoLayerRAG(; topK, reranker)
 end
diff --git a/src/rerankers/PromptOptimizer.jl b/src/rerankers/PromptOptimizer.jl
@@ -1,5 +1,3 @@
-
-using PromptingTools
 using Dates
 
 export optimize_reranker, analyze_missing_targets
diff --git a/src/rerankers/ReduceGPTReranker.jl b/src/rerankers/ReduceGPTReranker.jl
@@ -1,8 +1,6 @@
-using PromptingTools
 using RAGTools: AbstractReranker
 using Base.Threads
 using HTTP.Exceptions: TimeoutError
-const PT = PromptingTools
 
 include("RankingExtractor.jl")  # Include our local extract_ranking function
 
@@ -43,15 +41,15 @@ function rerank(
 
             # Handle both string and vector prompts
             prompt = if prompt_result isa AbstractString
-                !isnothing(query_images) ? [PT.UserMessageWithImages(prompt_result; image_url=query_images)] : prompt_result
+                !isnothing(query_images) ? [UserMessage(; content=prompt_result, image_data=query_images)] : prompt_result
             elseif prompt_result isa Vector && !isnothing(query_images)
-                vcat(prompt_result[1:end-1], [PT.UserMessageWithImages(prompt_result[end].content; image_url=query_images)])
+                vcat(prompt_result[1:end-1], [UserMessage(; content=prompt_result[end].content, image_data=query_images)])
             else
                 prompt_result
             end
 
             response = try_generate(ai_manager, prompt; 
-                api_kwargs=(; temperature, top_p=0.1),
+                api_kwargs=(; temperature, ), # top_p=0.1
                 verbose=false,
                 request_id
             )
diff --git a/src/transform/LLM_CodeCriticsArchitect.jl b/src/transform/LLM_CodeCriticsArchitect.jl
@@ -1,8 +1,6 @@
 # ./src/transform/LLM_CodeCriticsArchitect.jl
-using PromptingTools: SystemMessage, UserMessage
 using Markdown
-using StreamCallbacksExt
-using StreamCallbacksExt: format_ai_meta, format_user_meta
+using OpenRouter: format_ai_meta, format_user_meta
 
 export CodeCriticsArchitectContext, LLM_CodeCriticsArchitect
 
@@ -46,10 +44,10 @@ function transform(ctx::CodeCriticsArchitectContext, query, session::Session; io
     if ctx.model == "o3m" # NOTE: o3m does not support temperature and top_p
         api_kwargs = (; )
     end
-    response = aigenerate([
+    response = aigen([
         SystemMessage(CRITICS_SYSTEM_PROMPT),
         UserMessage(prompt)
-    ], model=ctx.model, api_kwargs=api_kwargs, streamcallback=cb)
+    ], ctx.model; streamcallback=cb, api_kwargs...)
     
     content = response.content
     display(ctx, content)
diff --git a/src/transform/LLM_ExecutionPlanner.jl b/src/transform/LLM_ExecutionPlanner.jl
diff --git a/src/transform/LLM_apply_changes.jl b/src/transform/LLM_apply_changes.jl
diff --git a/src/transform/LLM_reflect.jl b/src/transform/LLM_reflect.jl
diff --git a/src/transform/StreamingHighlight.jl b/src/transform/StreamingHighlight.jl
diff --git a/src/transform/classifier.jl b/src/transform/classifier.jl
diff --git a/src/utils/AIGenerateFallback.jl b/src/utils/AIGenerateFallback.jl
diff --git a/src/utils/APIKeyManager.jl b/src/utils/APIKeyManager.jl
diff --git a/src/utils/APIKeyRotation.jl b/src/utils/APIKeyRotation.jl
diff --git a/src/utils/ModelConfig.jl b/src/utils/ModelConfig.jl
diff --git a/test/agents/test_fluid_agent.jl b/test/agents/test_fluid_agent.jl
diff --git a/test/test_api_key_manager.jl b/test/test_api_key_manager.jl
diff --git a/test/test_api_key_rotation.jl b/test/test_api_key_rotation.jl
diff --git a/test/test_model_config.jl b/test/test_model_config.jl

Original file line number	Diff line number	Diff line change
`@@ -1,4 +1,3 @@`
`1`		`-using PromptingTools`
`2`	`1`
`3`	`2`	`"""`
`4`	`3`	`condition_llm(condition::AbstractString; model::AbstractString="gpt4om", verbose::Bool=false)`
Original file line number	Diff line number	Diff line change
`@@ -1,4 +1,3 @@`
`1`		`-using PromptingTools`
`2`	`1`
`3`	`2`	`export selector_llm`
`4`	`3`
Original file line number	Diff line number	Diff line change
`@@ -1,5 +1,4 @@`
`1`	`1`	`using EasyContext`
`2`		`-using PromptingTools`
`3`	`2`
`4`	`3`	`"""`
`5`	`4`	`This example demonstrates various ways to use FluidAgent for automation tasks.`
Original file line number	Diff line number	Diff line change
`@@ -1,4 +1,4 @@`
`1`		`-using PromptingTools: recursive_splitter`
	`1`	`+import PromptingTools: recursive_splitter`
`2`	`2`	`using RAGTools`
`3`	`3`	`using RAGTools: AbstractChunker`
`4`	`4`