From 4d20d36bfd8f5c3c6631726f29f6f528a029752e Mon Sep 17 00:00:00 2001
From: Jonas Pacheco <jonaspachecoometas@github.com>
Date: Mon, 23 Mar 2026 13:02:57 -0300
Subject: [PATCH] feat(ai): adiciona DeepSeek R1 14B como agente principal no
 LiteLLM
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- Adiciona deepseek-r1:14b e alias arcadia-agent no litellm-config.yaml
- arcadia-default agora aponta para R1 14B (era llama3.2:3b)
- Timeout LiteLLM aumentado de 120s para 300s
- Adiciona rede ollama-net (external) ao serviço litellm no compose
  para comunicação com container Ollama via hostname estável
- .gitignore: ignora arquivos .aider*

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---
 .gitignore                 |  1 +
 docker-compose.prod.yml    |  4 ++++
 docker/litellm-config.yaml | 26 ++++++++++++++++++--------
 3 files changed, 23 insertions(+), 8 deletions(-)

diff --git a/.gitignore b/.gitignore
index c4973d3..cbb8d9b 100644
--- a/.gitignore
+++ b/.gitignore
@@ -77,3 +77,4 @@ REQUISITOS_*.md
 *.txt
 !README.md
 !replit.md
+.aider*
diff --git a/docker-compose.prod.yml b/docker-compose.prod.yml
index 819b151..7c17b0f 100644
--- a/docker-compose.prod.yml
+++ b/docker-compose.prod.yml
@@ -305,6 +305,7 @@ services:
     networks:
       - arcadia-internal
       - coolify
+      - ollama-net
 
   # ── Ollama (LLMs locais — soberania total) ────────────────────────────────────
   # OPÇÃO A (padrão): Ollama como container Docker
@@ -393,6 +394,9 @@ networks:
     driver: bridge
   coolify:
     external: true
+  ollama-net:
+    external: true
+    name: ia1upsekrad96at5hq97e4qa
 
 volumes:
   pgdata:
diff --git a/docker/litellm-config.yaml b/docker/litellm-config.yaml
index 0085857..4de09a5 100644
--- a/docker/litellm-config.yaml
+++ b/docker/litellm-config.yaml
@@ -31,6 +31,16 @@ model_list:
       model: ollama/llama3.2:3b
       api_base: os.environ/OLLAMA_BASE_URL
 
+  - model_name: deepseek-r1
+    litellm_params:
+      model: ollama/deepseek-r1:14b
+      api_base: os.environ/OLLAMA_BASE_URL
+
+  - model_name: arcadia-agent
+    litellm_params:
+      model: ollama/deepseek-r1:14b
+      api_base: os.environ/OLLAMA_BASE_URL
+
   - model_name: nomic-embed-text
     litellm_params:
       model: ollama/nomic-embed-text
@@ -66,25 +76,25 @@ model_list:
   #     api_key: os.environ/GROQ_API_KEY
 
   # ── Modelo padrão do Arcádia (Manus usa este) ─────────────────────────────────
-  # Prioridade: LLMFit (TIER 1) → Ollama (TIER 2 — fallback)
+  # Prioridade: DeepSeek R1 14B (TIER 2) → llama3.2 (fallback rápido)
   - model_name: arcadia-default
     litellm_params:
-      model: openai/llama3.2:3b
-      api_base: os.environ/LLMFIT_BASE_URL
-      api_key: os.environ/LLMFIT_API_KEY
+      model: ollama/deepseek-r1:14b
+      api_base: os.environ/OLLAMA_BASE_URL
 
 router_settings:
   routing_strategy: least-busy
   fallbacks:
-    - {"gpt-4o": ["llama3.2"]}
-    - {"gpt-4o-mini": ["llama3.2"]}
+    - {"gpt-4o": ["deepseek-r1"]}
+    - {"gpt-4o-mini": ["deepseek-r1"]}
     - {"arcadia-default": ["llama3.2"]}
-    - {"arcadia-finetuned": ["llama3.2"]}
+    - {"arcadia-agent": ["llama3.2"]}
+    - {"arcadia-finetuned": ["deepseek-r1"]}
     - {"arcadia-embed": ["nomic-embed-text"]}
 
 litellm_settings:
   drop_params: true
-  request_timeout: 120
+  request_timeout: 300
   set_verbose: false
 
 general_settings: