Compiled by Amit KumarEdgee is an agent gateway from Edgee Cloud SAS, a Paris company with a US entity, Edgee Corporation. It sits between coding agents such as Claude Code, Codex, Cursor and GitHub Copilot and the model providers they call. It reroutes requests to cheaper models as a token budget fills or a provider fails, compresses tokens, and reports cost per developer, model, repo and pull request. Free costs $0 for one developer. Team is $30 per developer a month. Enterprise is priced by sales, and your own provider keys carry no commission on any plan.
Our read: this is a product for engineering leads whose agent bill has become a line item nobody can explain. The Free plan is generous enough to try on one machine, and the vendor is unusually frank about its savings claim, publishing a production figure of 15 to 20% and telling you not to add it to its 50% benchmark figure. The catches are practical. Team seats include no tokens, Free adds a 5% fee to Edgee credit top-ups, and the vendor's own pages disagree on whether routing strategies need a paid seat. We read the Edgee homepage, the Edgee pricing page, the Edgee Mod for Claude Code page, the documentation, the Terms of Use and the Privacy Policy on 11 October 2026.
What does Edgee do?
Edgee works as a proxy for the agents your developers already run. You install the Edgee CLI and start an agent through it, for example with edgee launch claude, and its model traffic then passes through Edgee's gateway. The agent itself does not change.
- Budget-driven routing. A routing Strategy pairs a token budget with rules that fire as it fills. The routing page gives an example: on a $100 weekly squad budget, Opus-bound requests move to another model at 75%, Sonnet-bound ones at 85%, and everything goes to a cheaper floor model at 100% instead of being blocked.
- Fallback. On a rate limit, server error or exhausted plan cap, Edgee retries on the next model in the chain without restarting the session.
- Token compression. Three switchable techniques trim tool results, shrink the list of tools sent to the model and keep model answers short. The compression page quotes two numbers and asks you not to add them: 15 to 20% lower token bills on live customer fleets over a rolling 30 days, and a 50% cut on the SWE-bench Lite benchmark with all three techniques on.
- Observability. Tokens, cost and latency per session, developer and model; on Team, per repo and pull request via GitHub.
- Plugins. Admins package skills and MCP servers once and Edgee installs them in each agent, per the plugins page.
- Bring your own keys. You can keep paying Anthropic, OpenAI or another provider directly with your own keys or subscription, or buy tokens from Edgee.
Edgee Mod for Claude Code
The newest piece is the Edgee Mod. The Edgee changelog entry of 8 October 2026 describes a sidebar inside Claude Code where you search 200+ models, reroute the current session to one of them and see estimated compression and routing savings live. A chosen model applies to the whole session for 24 hours, and "Claude Code's choice" switches back. It ships with the CLI and needs Claude Code 2.1.287 or newer. Its savings figures are estimates, and the console stays the reference for spend.
The coding agents page lists Claude Code, Codex, OpenCode, Cursor, GitHub Copilot, Xcode and others; desktop apps are in beta.
Who is Edgee for, and who should skip it?
Edgee fits engineering leads who cannot say which developer or repo spent what. Per-developer and per-PR attribution is the main reason to look. It also suits teams that hit plan caps mid-sprint and want a fallback rather than a stalled session.
Skip it in these cases:
- You are one developer on a flat subscription who never hits limits.
- Your company forbids sending prompts through a third-party proxy and you cannot buy the on-premise licence, which is sold through the Enterprise plan.
- You want a plain multi-model API for an application you control. Edgee's own comparison with OpenRouter says it is built around wrapping agents, not around a single endpoint for apps.
How much does Edgee cost?
USD, checked 11 October 2026, from Edgee pricing
| Plan | Price | What you get | Where it stops |
|---|---|---|---|
| Free | $0, no credit card | One developer, all their coding agents, rerouting strategies, input and output token compression, personal dashboard for usage, cost and sessions, BYOK with 0% commission, community and email support | One member only; 5% fee on Edgee credit top-ups; no smart routing, team dashboards, plugins, spending caps or GitHub attribution |
| Team | $30 per developer per month | Everything in Free plus smart routing by task complexity (beta), team and squad observability, plugins, data destinations, model selection, GitHub per-repo and per-PR attribution, spending caps and alerts, priority support; 0% top-up fee | Seats do not include tokens; no SSO, private gateway or contractual SLA |
| Enterprise | Custom (contact sales) | Everything in Team plus SSO/SAML, private gateway as SaaS or on-premise, gateway residency in the EU, US or APAC, privacy controls such as PII redaction, private open-weight models paid per GPU hour, 30-minute SLA on urgent tickets | Price not published; credits billed on a dedicated invoice, payable later |
Billing has two parts, and the pricing FAQ is clear that they are separate. Seats pay for Edgee's features. Tokens are paid either to your provider, when you use your own key or subscription, or to Edgee as prepaid credits charged at provider prices. On Free, adding $100 of credits costs $105 before tax; on Team it costs $100.
One more price sits outside the pricing table. The open-weight models page offers "Turbo" versions of models such as GLM 5.2, Kimi K3 and MiniMax 2.7 in Claude Code for a flat $29 a month, with no per-token metering. The pricing page does not list it, so confirm how it combines with Free or Team.
Is Edgee free?
For one developer, yes. Free costs $0, needs no credit card and includes compression, rerouting strategies and a personal dashboard. Tokens are still paid for, to your provider or as Edgee credits with the 5% top-up fee. One wrinkle: the pricing page lists rerouting strategies on Free, while the routing page says "Team plan required for routing strategies". Test budget rules before you count on them without a seat.
Is Edgee worth it for a small team?
Run the arithmetic on your own bill. A Team seat is $30 per developer a month. If compression alone delivers the 15 to 20% the vendor reports from production, a developer would need to spend roughly $150 to $200 a month on tokens before compression covers the seat, before any routing savings. Below that, the case rests on attribution, caps and fallback rather than savings. The seat price and the percentage are Edgee's; the break-even is our arithmetic.
Does Edgee store my prompts or code?
Not by default, according to the privacy policy. It says prompts and completions are processed in memory only and are not stored unless a customer turns on debug mode, in which case they are kept for up to 30 days. Usage metadata such as token counts, latency and model is kept for up to 25 months. Operational data stays in the EU by default; named subprocessors are AWS, Vercel and ClickHouse, plus Stripe for payments. The policy states that Edgee never uses customer prompts, completions or usage metadata to train or fine-tune models. Prompts still reach the provider your route selects. The footer links SOC 2 and GDPR pages; ask Edgee for the audit report if you need it.
The on-premise page offers the gateway as a container (Docker Compose or Helm, connected or air-gapped) so prompts and keys stay on your network; the FAQ calls it a licensed, proprietary deployment. The CLI is open source under Apache 2.0 on Edgee's GitHub repository.
Who makes Edgee?
The terms and privacy policy name Edgee Cloud SAS, based in Paris (75002), and its US entity Edgee Corporation. The terms are governed by French law, with disputes going to the courts of Paris. Tech.eu reported on 17 October 2024 that founders Sacha Morard, a former CTO of Le Monde Group, and Gilles Raymond raised $2.9 million in pre-seed funding from Serena and VentureFriends; at the time Edgee was an edge computing platform for data collection. The Edgee Mod for Claude Code was featured on Product Hunt on 11 October 2026.
How Edgee compares with other tools reviewed here
Three other tools we have reviewed in the development category also sit between a developer and an AI agent, each with a different job.
- VibeDefend for Claude Code, Cursor and Copilot is a security layer: it injects rules and scans the diff while the agent writes code. Our review lists a free plan at $0 with no card and a Business plan priced after a pilot, checked 13 October 2026. VibeDefend stops risky code; Edgee controls what the tokens cost.
- CoreSpeed's MCP access layer gives agents one endpoint for connected apps, memory and spend caps. Our review lists a Free start with 3,000 credits for 90 days and Pro at $20 a month, checked 4 October 2026. CoreSpeed governs which tools an agent reaches; Edgee governs which model answers.
- Bolt.new and its token plans is an app builder where the vendor meters tokens for you: 1M tokens a month free and Pro at $25 a month, per our check on 24 September 2026. Edgee is a meter you put on agents you already pay for elsewhere.
Sources
Plans, seat prices, the credit top-up fee, BYOK terms and the pricing FAQ come from the Edgee pricing page. Routing behaviour and the budget example come from the smart routing and budgets page, compression figures and methodology from the token compression page, and the Turbo plan from the open-weight models page. The Claude Code sidebar is described on the Edgee Mod page and dated in the Edgee changelog. Data handling comes from the Edgee Privacy Policy and the Edgee documentation FAQ, the company and governing law from the Edgee Terms of Use, and the founders and funding from Tech.eu's pre-seed report. Vendor pages were read on 11 October 2026.
