Edgee is a transparent proxy that sits between AI coding assistants and LLM providers, compressing input and output tokens to cut usage costs by up to 50% without any code changes. It offers a drop‑in CLI, dual‑layer token compression, automatic provider routing with retry logic, and real‑time observability dashboards, supporting a wide range of coding agents and OpenAI‑compatible APIs.
Funding
Funding not disclosed
Founders
Product
Problem
AI-powered coding assistants generate large prompts and responses, leading to high token consumption and increased costs for developers and organizations. Managing token usage often requires modifying code or adding custom middleware, which adds complexity and delays adoption.
Solution
Edgee provides an edge intelligence layer that sits between coding agents and LLM providers, acting as a transparent proxy that compresses both input and output tokens without any code changes. It applies two compression layers—tool result trimming and tool surface reduction for inputs, and output brevity for responses—delivering up to 50% token cost reduction while preserving code correctness. The platform also routes requests with automatic fallback and retry logic, ensuring reliability across multiple LLM providers. Detailed observability dashboards report per‑session compression ratios, token savings, and cost avoidance, allowing teams to monitor and optimize usage in real time. Edgee supports a wide range of coding agents (Claude Code, Codex, Cursor, etc.) and any OpenAI‑compatible provider through BYOK API keys, making integration simple and provider‑agnostic.
Target Audience
Edgee is aimed at software development teams, DevOps engineers, and enterprises that integrate AI coding assistants into their development workflows and need to control LLM token costs without altering existing pipelines.
Features
- Drop‑in CLI wrapper that works as a transparent proxy; no modifications to existing coding agents required
- Dual‑layer token compression: input tool result trimming (60–90% reduction) and tool surface reduction, plus configurable output brevity levels
- Automatic provider routing with retry and fallback based on real‑time success rates, supporting Anthropic, OpenAI, Google Vertex AI, Mistral, DeepSeek, xAI, AWS Bedrock, and others via BYOK
- Real‑time observability console showing token usage, compression savings, latency, and error logs at session and team levels
- SDKs for TypeScript, Python, Go, and Rust with tag support for granular analytics and cost tracking
- Sub‑15 ms edge compression latency, ensuring minimal impact on response times