Added CrewAI, Gradio MCP server, and Modal integration

agents.yaml

@@ -0,0 +1,24 @@
+- role: Repository Cloner Agent
+  goal: Clone the GitHub repository.
+  backstory: The fastest fetcher of code.
+  tools: [repository_cloner_agent]
+
+- role: Code Analysis Agent
+  goal: Analyze code for issues and improvements.
+  backstory: A bug hunter with sharp eyes.
+  tools: [code_analysis_agent]
+
+- role: Metadata Agent
+  goal: Extract lines of code, languages, stars.
+  backstory: A stats wizard.
+  tools: [metadata_agent]
+
+- role: Documentation Agent
+  goal: Review README and docstrings.
+  backstory: A language wizard.
+  tools: [documentation_agent]
+
+- role: Orchestrator Agent
+  goal: Combine all findings into a final report.
+  backstory: The conductor of the orchestra.
+  tools: [orchestrator_agent]

app.py

@@ -0,0 +1,16 @@
+import gradio as gr
+from crew import run_crew
+
+def process_query(user_input):
+    return run_crew(user_input)
+
+iface = gr.Interface(
+    fn=process_query,
+    inputs=gr.Textbox(label="Enter GitHub repo URL for analysis"),
+    outputs=gr.Textbox(label="Analysis report"),
+    title="🎶 Distributed Mistral Orchestra",
+    description="Analyze a GitHub repo using a multi-agent AI team powered by Mistral models on Modal."
+)
+
+if __name__ == "__main__":
+    iface.launch(mcp_server=True)

crew.py

@@ -0,0 +1,12 @@
+from crewai import Crew
+
+crew = Crew.from_yaml(
+    agents_file='agents.yaml',
+    tasks_file='tasks.yaml',
+    process='sequential'
+)
+
+def run_crew(github_url):
+    for task in crew.tasks:
+        task.description += f"\nUser GitHub URL: {github_url}"
+    return crew.kickoff()

modal_agents.py

@@ -0,0 +1,19 @@
+import modal
+
+client = modal.Client()
+app = client.get("distributed-mistral-orchestra")
+
+def repository_cloner_agent(github_url):
+    return app.repository_cloner_agent.call(github_url)
+
+def code_analysis_agent(prompt):
+    return app.code_analysis_agent.call(prompt)
+
+def metadata_agent(prompt):
+    return app.metadata_agent.call(prompt)
+
+def documentation_agent(prompt):
+    return app.documentation_agent.call(prompt)
+
+def orchestrator_agent(summary):
+    return app.orchestrator_agent.call(summary)

modal_app.py

@@ -0,0 +1,44 @@
+import modal
+
+app = modal.App("distributed-mistral-orchestra")  # <--- app, not stub
+@app.function(gpu="A100")
+def repository_cloner_agent(github_url):
+    import subprocess
+    subprocess.run(["git", "clone", github_url, "/root/repo"], check=True)
+    return f"Cloned repository: {github_url}"
+
+@app.function(gpu="A100")
+def code_analysis_agent(prompt):
+    from transformers import AutoTokenizer, AutoModelForCausalLM
+    tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-Large")
+    model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-Large").to("cuda")
+    inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
+    outputs = model.generate(**inputs, max_new_tokens=300)
+    return tokenizer.decode(outputs[0], skip_special_tokens=True)
+
+@app.function(gpu="H100")
+def metadata_agent(prompt):
+    from transformers import AutoTokenizer, AutoModelForCausalLM
+    tokenizer = AutoTokenizer.from_pretrained("mistralai/Mixtral-8x7B-Instruct-v0.1")
+    model = AutoModelForCausalLM.from_pretrained("mistralai/Mixtral-8x7B-Instruct-v0.1").to("cuda")
+    inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
+    outputs = model.generate(**inputs, max_new_tokens=250)
+    return tokenizer.decode(outputs[0], skip_special_tokens=True)
+
+@app.function(gpu="T4")
+def documentation_agent(prompt):
+    from transformers import AutoTokenizer, AutoModelForCausalLM
+    tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2")
+    model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-Instruct-v0.2").to("cuda")
+    inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
+    outputs = model.generate(**inputs, max_new_tokens=200)
+    return tokenizer.decode(outputs[0], skip_special_tokens=True)
+
+@app.function(gpu="A100")
+def orchestrator_agent(summary):
+    from transformers import AutoTokenizer, AutoModelForCausalLM
+    tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-Large")
+    model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-Large").to("cuda")
+    inputs = tokenizer(summary, return_tensors="pt").to("cuda")
+    outputs = model.generate(**inputs, max_new_tokens=400)
+    return tokenizer.decode(outputs[0], skip_special_tokens=True)

requirements.txt

@@ -0,0 +1,4 @@
+gradio[mcp]
+crewai
+requests
+modal

tasks.yaml

@@ -0,0 +1,19 @@
+- description: Clone the user's GitHub repo.
+  expected_output: Repo cloned for analysis.
+  agent: Repository Cloner Agent
+
+- description: Analyze the codebase for bugs and improvements.
+  expected_output: Code analysis report.
+  agent: Code Analysis Agent
+
+- description: Extract metadata like lines of code and languages.
+  expected_output: Repo metadata report.
+  agent: Metadata Agent
+
+- description: Review documentation for clarity.
+  expected_output: Documentation analysis.
+  agent: Documentation Agent
+
+- description: Compile all findings into a final report.
+  expected_output: Comprehensive analysis report.
+  agent: Orchestrator Agent