Chat with every model.
Outlast any limit.
One chat in VS Code, every model on one chain. When a provider caps, the next model picks up the same thread — and you see exactly what carried over. It runs on the subscriptions you already pay for.
One chat.
That keeps going.
A chat panel inside VS Code where Claude (up to Fable 5), GPT, Gemini and 19 free models share one conversation. When one caps, the next continues the same thread.
A model caps and greys out. The pulse carries your thread to the next one, with no button to press — and you can see what carried over.
One month on one machine, counted locally and priced at paid rates.
▶ it's playing live at the top of this page — scroll up any time
Order your models — a primary, then fallbacks, ending in free open-source ones. A limit on one exhausts its whole provider (their models share quota), so it jumps to the next one with tokens left. Preview the whole path anytime — no request, no quota.
Because the failover chain ends in free models — 22 of them via one OpenRouter key: Qwen3 Coder, Nemotron 550B, Llama, GPT-OSS, Gemma, Hermes 405B and more — the thread keeps going after your paid models cap. Free models have their own daily limits too — so Renitor spreads across different hosts and routes around the congested ones automatically. The switch is automatic — no button, no file, and you can see what carried over.
On the latest reply, ask another model of your choice to critique it, validate it, or suggest what comes next — right there. Send its take back to the chat with one click.
The receipts.
We told Claude a codeword, let it hit a limit, detoured through Codex, then resumed. Claude recalled the codeword and described the detour's commits — from its own memory.
Full transcript in the docs — reproduce it yourself with two tiny prompts.
How much longer can you actually work?
A single plan gives you a couple of hours, then blocks you for the rest of its window. Renitor stitches your plans into one chain: when Claude caps, Codex takes over; when that caps, Gemini and the free tail — and by the time you circle back, Claude's window has reset. Toggle what you pay for and watch a workday fill in:
An honest estimate, not a promise. Hours per window vary with how heavy each turn is — these assume steady coding on the entry plans (Claude Pro / ChatGPT Plus ~2 hrs per 5-hour window; Gemini & the free tier ~30 min/day each). The point isn't the exact number — it's that a single plan stops, and a chain of independent quota pools keeps going. The real figure is the one Renitor counts on your machine: limits survived.
A full workspace, not a wrapper.
Memory you allocate
A per-project memory file every model can share — and you decide which models get to read it. Free models can help without seeing everything.
Sessions, themes, voice
Named sessions that survive restarts; faithful ChatGPT / Claude / Gemini themes in light & dark; a mic that transcribes straight into the composer.
Local-first, documented network behaviour
No backend, no telemetry, no phone-home to Renitor. Handoffs make no network calls; chat goes only to the provider you pick. The Pro license verifies offline, and your auth tokens are never touched.
Any chat LLM, too
ChatGPT free, DeepSeek, Gemini web — chat mode inlines the whole brief, because web chats can't read files.
Goal auto-inference
Reads the local session logs of four agents to fill in what you were doing — your latest instruction, not a stale one.
Secrets scrubbed twice
Provider-key patterns plus literal values learned from your local .env — redacted anywhere they appear.
Failing commands, captured
Non-zero exits are recorded automatically and land in the next handoff's error section. No copy-paste.
PR & focused handoffs
Brief the next agent on a whole branch vs base, or exactly the files you select in the Explorer.
You already pay for enough intelligence.
Renitor makes it act as one.
A Claude sub here, a GPT one there, free tiers, maybe a Gemini key. You are rich in models and poor in continuity. Renitor is built on one conviction: everything you already pay for should behave like a single, unlimited, honest AI workspace.
Your models, your order, saved per project. Renitor routes around limits automatically, and it never chooses for you.
Merge them (Consensus), pit them (Debate), employ them (Boss & Worker), hang out (Group). One door per job.
Local-first, zero telemetry, offline license. Savings are counted on your machine. Every failure names its next move, and every claim on this page reproduces on a real CLI.
The next reset is coming.
Be working through it.
Install in two minutes. Everything is unlocked.