How to connect Reality Router to Cline

One dropdown in Cline's settings. On a real 612-request stretch, RR cut the bill from $77.30 to $9.80 — same tasks, 87% off. Bonus for Cline specifically: RR automatically handles the "big model for Plan mode, fast model for Act mode" split you were about to configure by hand.

If you're using Cline in VS Code — the fastest-growing BYOK coding agent out there — you're already the kind of dev who wanted control over which model runs your agent. Pointing Cline at Reality Router keeps every ounce of that control and adds four things:

• Per-call model choice, automatically. RealitySignal calibrated probabilities pick the right model per step. No more manually toggling between Haiku and Opus mid-session.
• Plan/Act split without the config work. Cline's killer feature is letting you pick different models for Plan vs Act. With RR + auto, that split happens on every call, not just at the mode boundary.
• A dashboard that outlives the session. Cline shows per-task cost in the sidebar; RR aggregates every task across every project into one view.
• No provider caps. Cline is BYOK, which means your caps come from wherever your keys point. Route through RR and those become RR's per-token pricing — no hourly limits, no weekly resets.

Here's the setup.

1. Add RR as a provider (30 seconds)

• Click the ⚙️ settings icon in the Cline panel
• Open the API Provider dropdown at the top
• Select OpenAI Compatible
• Fill in:
  • Base URL: http://localhost:8000/v1 (or http://<your-host>:8000/v1 if RR is on a different box)
  • API Key: rr-local (RR accepts anything for local instances — the header just needs to exist)
  • Model: auto (RR's RealitySignal picks per call. No prefix required — Cline sends the model ID verbatim.)
• In the Advanced section, tick:
  • Image Support — enable if you want vision-capable models (Opus 5, GPT-5) to receive images from Cline's screenshots
  • Computer Use — enable so Cline can call tools via the routed provider
• Click Verify. Green check = you're live.

2. Bonus: use RR for both Plan and Act (or split them)

Cline's Plan/Act mode lets you pick a different model for each. Two ways to combine with RR:

Option A (recommended — simplest): Leave "Use different models for Plan and Act" off. Both modes use auto. RR routes each call individually — Plan-mode calls tend to be reasoning-heavy so RealitySignal picks stronger models; Act-mode calls tend to be short & mechanical so it picks cheaper ones. The split you'd have configured manually happens automatically.

Option B (if you want explicit control): Enable "Use different models for Plan and Act." Pin:

• Plan mode: claude-opus-5 (or gpt-5) via RR — RR still records the call, tracks cost, and can fall back if the provider errors
• Act mode: claude-haiku-4-5 (or deepseek-v4-flash) via RR

Both routes go through the same dashboard.

3. Verify it worked (10 seconds)

Ask Cline anything trivial:

Explain what this project does.

Then open the RR dashboard at http://localhost:8000. Your Cline call shows up in the Agent Activity table.

What you'll see

Dashboard mid-task, Cline at the top:

Enlarge

Total Volume: 612 requests. Accrued Expense: $9.80. Potential Cost (if everything had run on the top model in the pool): $77.30. Retained Value: $67.50 — 87% of the bill you'd have paid without routing, saved automatically. The top row of Agent Activity is Cline itself.

Why bother

Cline is the BYOK agent. You picked it because you wanted to see the API key, pick the model, control the cost. That's exactly the mindset RR is built for.

Without RR, that control comes with homework: pick the right model for the right task, remember to swap between them, manually configure Plan vs Act, watch your Anthropic and OpenAI dashboards separately, guess whether that failed call was rate-limit or bug. Every one of those decisions is a distraction from what you actually wanted to do — ship code.

Route Cline through RR and:

• One provider config replaces every per-vendor key. Anthropic key, OpenAI key, DeepSeek key, Groq key — all handed to RR once, at the server. Cline sees one endpoint.
• Runaway loops become impossible to hide. Cline's own cost tally is per-task; RR's dashboard is cross-task, real-time, and alerts on anomalies. A misconfigured Act mode that keeps re-editing the same file surfaces immediately.
• No caps to hit. Kick off a long-running agent job — a full codebase audit, a batch refactor, a research crawl — RR bills per token, no bucket, no reset window.

Same VS Code. Same Cline panel. Same Plan/Act. Just cheaper, safer, and measured.

→ realityrouter.dev

Reality Router is open source and self-hosted. Metrics in this post are from a real 612-request run.