EcoAgent

• 

  Main Page

• 

  Demo result

• 

  Demo result

• 

  Demo result

• 

  Demo result

About the Project Stories about runaway AI bills, data exposure, and the environmental cost of computing inspired us to build EcoAgent. Many teams want to try AI models and agents but may not know how quickly usage can grow—or how to keep an agent within the limits of an assignment.

EcoAgent makes model routing easier to inspect. It compares a fixed large-model baseline with a backend-selected routing policy, showing the models and routing reasons, outputs, latency, and token usage. It also provides energy and carbon estimates based on disclosed assumptions, so users can explore possible tradeoffs without mistaking estimates for measured emissions. For local Ollama inference on supported NVIDIA hardware, EcoAgent can also show best-effort GPU readings for the whole device during a request; these readings are not attributed to a specific process or user.

We built the project with a React, TypeScript, and Vite frontend and a Python FastAPI backend. The prototype supports several model providers and includes a benchmark workflow and a guarded Workspace Agent with manager and specialist roles, explicit approvals, and stop controls.

One of our biggest challenges was making results understandable without overstating what they prove. Provider usage, local GPU telemetry, simulated demo data, and illustrative energy estimates have different levels of certainty, so we worked to label their sources and limitations clearly. We also learned that agent safety is more than preventing repeated actions: users need to see what the agent is doing and retain control over approvals and stopping work.

EcoAgent is a prototype for helping teams explore AI routing and agent safeguards. Its comparisons do not establish equal answer quality or prove real-world emissions reductions; those claims require broader evaluation and better-calibrated measurements.

Built With

• ai
• ai-inference
• anthropic-api
• benchmarking
• carbon-footprint
• energy-efficiency
• fastapi
• gemini-api
• groq
• llm
• model-routing
• openai-api
• python
• react
• sqlite
• sustainability
• typescript
• vite