Agent Control Plane

A platform to build, configure, and run AI agents — not a chatbot.

You define the agent. You define the tools. Claude does the rest.

Point any agent at your database. Give it tools to read, write, cancel, refund — whatever your domain needs. The e-commerce demo below is just one example of what you can build in minutes.

---

🤔 Why This Is Not Just Another Chatbot

Most chatbots are stateless text boxes wired to an LLM.

This is a configurable agent platform:

• Real tool execution — agents don't simulate actions, they run Python functions against a real database. Cancel an order? It's actually cancelled.
• Agentic loop — Claude decides which tool to call, reads the result, and can chain multiple tool calls before responding. Not one-shot, not hardcoded.
• Centralized tool registry — all tool calls flow through a single call_tool() entry point. This makes it trivial to add logging, auth checks, rate limiting, or usage tracking in one place — not scattered across every tool.
• Developer-first configuration — agents are defined at runtime via UI (model, system prompt, tools). No code changes to spin up a new agent.
• Built-in observability — every run logs input tokens, output tokens, and latency per API call. Token costs and response times are captured automatically because all calls go through one runner.
• Auto-generated tool schemas — add @register_tool to any Python function and it's instantly available to any agent. Anthropic-compatible JSON schema is generated from type hints — no manual JSON writing.

👇 See the Demo section below for screenshots of the chat, transcript, configuration, and usage dashboard in action.

---

🛍️ Demo — E-commerce Customer Support Agent

The included demo configures an agent with real database access to handle customer support queries — looking up orders, cancelling orders, requesting refunds, and updating shipping addresses.

Agent Configuration

Configure any agent via UI — set the model (Opus, Sonnet, Haiku), write a system prompt, and assign tools from the registered tool library.

Chat Transcript (with tool calls visible)

The history view shows the full turn-by-turn transcript including which tools were called, what they returned, and how Claude used the results to respond.

Usage & Metrics

Per-run logs track model used, input tokens, output tokens, and latency. Aggregate stats (total runs, total tokens, average latency) shown per agent.

---

⚙️ How It Works

Developer configures an agent (model + system prompt + tools) via UI
         ↓
User sends a message via chat
         ↓
Backend runs the agentic loop (Anthropic SDK)
  → Claude decides which tool to call
  → Tool executes against real DB
  → Result returned to Claude
  → Claude responds naturally
         ↓
Full turn-by-turn history + token/latency logs saved to DB

---

🗂️ Project Structure

backend/
  agents/        # Agent + Tool models, CRUD API
  chat/          # Session, Message, UsageLog — agentic loop
  tools/         # @register_tool registry + schema auto-generation
  ecommerce/     # Demo app — Customer, Order, OrderItem models + seed data
frontend/        # React UI

---

🛠️ Stack

Layer	Tech
Backend	Django 5.0 + Django REST Framework
Database	PostgreSQL
AI	Anthropic Python SDK (claude-sonnet-4-6)
Frontend	React

---

📖 See backend/readme.md for setup instructions and full API reference.