SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsOther AI toolsEdgee homepage

Edgee homepage

Edgee is an Agent Gateway for AI agents, used primarily for budget-driven model routing, token compression, and team usage observability. It sits between your agents and LLM providers and applies three major capabilities to every request: setting token budgets by developer, group, or entire organization, and when a budget is consumed to a certain extent, it can reroute requests mid-session to open-weight models, allegedly without quality loss and without interrupting the session; trimming input tool results and streamlining output to reduce billed tokens in a semantically lossless way; and observing cost and usage by developer, repository, PR, and model. It supports coding agents such as Claude Code, Codex, Cursor, and Copilot, requires no code changes, and can begin routing in about 5 minutes after installing the CLI and connecting. Edgee claims it can reduce costs by up to 70%, supports bring-your-own keys, and supports localized deployment.

Open website

About this tool

Edgee is an Agent Gateway positioned between AI agents and LLM providers, uniformly applying three capabilities to every request: budget-driven routing, token compression, and team observability. It officially claims it can reduce token costs by up to 70%, with no code changes required.

Purpose Edgee primarily serves two types of users: teams running coding agents, and teams delivering AI applications. It takes over routing, metering, and compression, and the company says onboarding takes about 5 minutes. Its philosophy is: in a model-driven world, intelligence has a cost, and tokens flow through every interaction, decision, and response, so the flow of intelligence should be compressed, routed, and optimized.

Main capabilities

  1. Budget-driven routing: token budgets can be set for each developer, each group, or the entire organization. As budgets gradually fill up, Edgee reroutes requests to open-weight models while tasks are in progress, and it officially says this causes no quality loss and does not interrupt sessions. It is also resilient: when a provider fails or a plan cap is triggered, routing continues along the next model in the policy.
  2. Token compression: the first layer (input) trims tool results, and the second layer (output) streamlines them. Reducing the payload of tool results at the gateway is semantically lossless for coding tasks: the same model output, fewer billed tokens.
  3. Team observability: view cost and usage by developer, by repository, by PR, and by model. Engineering leaders can see where spending goes before the bill arrives.
  4. Open-weight models: GLM, Kimi, and MiniMax are offered at a fixed monthly fee with turbo speed, serving as fallback targets in routing policies, and have been benchmarked for coding work.
  5. Bring your own keys: you can use Edgee's key for convenience, or connect your own provider keys to control billing and use custom models.
  6. Localized deployment: the same gateway can run within your own infrastructure. Prompts, completions, and provider keys do not leave your network. Docker Compose or Helm is supported, in connected or air-gapped environments.

How to use For coding agents: install the Edgee CLI and connect it to your coding agent, with no code changes required. Budgets can be set from the start, assigning routing policies to developers, groups, or the entire organization. It works as a transparent gateway and is compatible with Claude Code, Codex, Copilot, OpenCode, Cursor, and others. The company says connectingIt takes about 5 minutes to reach the coding agent and start routing.

Use cases

  • Teams use coding agents and want to control token spend and observe usage distribution.
  • Need to automatically switch to open-weight models when budget is exhausted or a provider fails to keep tasks continuous.
  • Want to reduce token billing caused by tool results and output at the gateway layer.
  • Organizations with data compliance requirements that need localized or air-gapped deployment.

Limitations and missing information The materials do not provide specific feature demonstrations, pricing details, a complete list of models, performance data, supported programming languages or platform details, nor do they give specific CLI installation commands. All effect descriptions (such as "up to 70% cost reduction" and "no quality loss") come from official statements, and independent verification information for actual performance is not yet available.

Related tools

Browse category