Connect Reality Router to your dev environment

One config change per tool. On a real 2,029-request stretch across five different coding agents, RR cut the total bill from $207.40 to $25.13 — same work, 88% off. Same tools, same keybindings, same workflow.

If your daily driver is one of the tools below, you're a single setting away from routing every call through Reality Router. No plugin, no fork, no changes to how you actually write code — just a smarter, cheaper, fully-metered version of what you already use.

The common pattern

Every tool in the list accepts a "custom OpenAI-compatible endpoint" — that's the plug RR was built for. Point it at http://localhost:8000/v1, drop in any placeholder API key, and RR does the rest:

• RealitySignal picks the model per call — real-time calibrated probability scoring, so cheap models handle boilerplate and expensive ones only fire when they'll actually help.
• Every request lands in a dashboard — model, tokens, cost, latency, success rate. Runaway loops and misconfigured routes surface within seconds, not at the end of the month.
• No caps to hit — pay per token, not per bucket. Kick off long agent jobs without staring at a fast-request meter or waiting for Monday's reset.

Here's what that looks like when all five tools route through the same RR instance at once:

Enlarge

Total Volume: 2,029 requests across five clients. Accrued Expense: $25.13. Potential Cost (if every call had run on the top model in the pool): $207.40. Retained Value: $182.27 saved automatically — 88% of the bill you'd have paid without routing.

The five setups

Each is a full walkthrough with a real dashboard receipt.

🖥 OpenCode

The reference setup — TUI agent, config-file driven. Add a provider block to ~/.config/opencode/opencode.json, done. ~1 minute. → How to connect Reality Router to OpenCode

🎨 Cursor

The biggest AI-IDE install base. Enable Override OpenAI Base URL in Settings → Models, point it at RR, click Verify. ~30 seconds. → How to connect Reality Router to Cursor

⚡ Aider

The cost-conscious dev's CLI. One flag: --openai-api-base http://localhost:8000/v1 --model openai/auto. Also works from .aider.conf.yml. ~30 seconds. → How to connect Reality Router to Aider

🔧 Cline (VS Code)

The fastest-growing BYOK agent. Settings gear → API Provider → OpenAI Compatible → fill in base URL. Bonus: RR handles the Plan/Act model split automatically. ~30 seconds. → How to connect Reality Router to Cline

🅰 Codex CLI

Yes, OpenAI's own CLI works with RR. Six lines of TOML in ~/.codex/config.toml, using the [model_providers.<name>] section OpenAI itself documents. ~1 minute. → How to connect Reality Router to Codex CLI

Your tool isn't listed?

Any tool that lets you set a custom OpenAI base URL works with RR. The OpenCode post has the most detailed walkthrough of the underlying pattern — apply it to whatever you're using. If your tool doesn't support custom endpoints, file an issue on GitHub and we'll help you work around it.

What you'll save (if you're weighing whether it's worth the setup)

Two case-study posts go deep on the numbers if you want the whole story:

• Cut the cord on your $200 AI subscription — how RR compares against ChatGPT Pro, Claude Max 20x, and OpenCode Go on the same budget
• Stop rationing your API budget — how RR compares against direct API calls to GPT-5, Fable 5, Haiku, and DeepSeek

Short version: on a $200 budget, RR runs ~7,140 tasks against Claude Max's ~1,300. Same work.

→ realityrouter.dev

Reality Router is open source and self-hosted. Metrics in this post are aggregated from the same real workload used in each per-tool how-to.