fix: preserve DeepSeek reasoning loops

- Add a native DeepSeek streaming transport for thinking tool use

- Persist assistant thinking while only replaying tool-call reasoning

- Refresh DeepSeek V4 model pricing and registry coverage

Co-Authored-By: devagent <devagent@egavrin>

Author: egavrin

models/deepseek.toml

@@ -2,13 +2,45 @@
 provider = "deepseek"
 base_url = "https://api.deepseek.com/v1"
 
+# Current V4 models. Pricing from https://api-docs.deepseek.com/quick_start/pricing.
+[deepseek-v4-flash]
+context_window = 1000000
+response_headroom = 384000
+default_max_tokens = 384000
+tokenizer = "deepseek"
+supports_tools = true
+supports_parallel_tools = false
+supports_temperature = false
+input_price_per_million = 0.14
+cache_hit_input_price_per_million = 0.028
+output_price_per_million = 0.28
+
+[deepseek-v4-pro]
+context_window = 1000000
+response_headroom = 384000
+default_max_tokens = 384000
+tokenizer = "deepseek"
+supports_tools = true
+supports_parallel_tools = false
+supports_temperature = false
+input_price_per_million = 1.74
+cache_hit_input_price_per_million = 0.145
+output_price_per_million = 3.48
+
+# Compatibility aliases. DeepSeek says these names will be deprecated in the future:
+# deepseek-chat maps to deepseek-v4-flash non-thinking mode, and deepseek-reasoner
+# maps to deepseek-v4-flash thinking mode.
 [deepseek-chat]
-context_window = 128000
-response_headroom = 4000
+context_window = 1000000
+response_headroom = 384000
+default_max_tokens = 384000
 tokenizer = "deepseek"
 supports_tools = true
 supports_parallel_tools = false
 supports_temperature = true
+input_price_per_million = 0.14
+cache_hit_input_price_per_million = 0.028
+output_price_per_million = 0.28
 
 [deepseek-coder]
 context_window = 128000
@@ -19,9 +51,13 @@ supports_parallel_tools = false
 supports_temperature = true
 
 [deepseek-reasoner]
-context_window = 64000
-response_headroom = 8000
+context_window = 1000000
+response_headroom = 384000
+default_max_tokens = 384000
 tokenizer = "deepseek"
 supports_tools = true
 supports_parallel_tools = false
 supports_temperature = false
+input_price_per_million = 0.14
+cache_hit_input_price_per_million = 0.028
+output_price_per_million = 0.28

packages/cli/src/main-session-persistence.ts

@@ -192,7 +192,7 @@ function registerSessionPersistenceEvents(bus: EventBus, sessionStore: SessionSt
   });
   bus.on("message:assistant", (event) => {
     const session = event.agentId || event.partial ? null : getActiveSession();
-    if (session) sessionStore.addMessage(session.id, { role: MessageRole.ASSISTANT, content: event.content, toolCalls: event.toolCalls });
+    if (session) sessionStore.addMessage(session.id, { role: MessageRole.ASSISTANT, content: event.content, thinking: event.thinking, toolCalls: event.toolCalls });
   });
   bus.on("message:tool", (event) => {
     const session = event.agentId ? null : getActiveSession();