Desktop app · your keys

Meerada LLManager

Every model, every conversation you've ever had — in one cockpit. Switch models mid-thought without losing a word.
Import your history from Claude Code, Claude.ai or ChatGPT and keep going on any model. Run many models on many tasks at once, on your own keys. Fork a conversation to a second model, attach your code folder, let a cheap model draft and a strong one polish, and have a judge pick the best answer — things no single vendor can offer, because we're not tied to one.
⬇ Download the app Try it in the browser →
Windows · macOS · Linux · your keys, your machine, your data. The engine underneath is open-source.
Import a Claude Code session, continue it on GPT-4o mini, then switch to GPT-4o mid-conversation — the whole history carried
30 seconds, real run: a Claude Code session imported → continued on GPT-4o mini → switched to GPT-4o mid-conversation. Nothing re-pasted, nothing lost.

🤝 The Handshake, built in

Your conversations belong to you, not to the model you happened to start them in. LLManager moves them freely — with the whole history.
📥

Import everything

One click finds your Claude Code sessions on this machine. Upload a Claude.ai or ChatGPT export, or paste any transcript. It becomes a live session — full history, on the model you choose.

↔

Switch mid-conversation

Change the model in the dropdown and the entire conversation moves with it — history, attached files, everything. Start in Claude, continue in GPT, finish in DeepSeek. Nothing is lost, nothing re-pasted.

⇄

Fork to a second model

Copy a conversation to another model and let both continue side by side. The same context, two opinions, one screen — the honest way to see who's better for your work.

📁

Attach files & folders

Point a session at your code folder (vendor/build dirs skipped, size-capped) or drop files in. It becomes the session's working context on every model it moves to.

⚡

Relay: draft cheap, polish strong

A cheap, fast model drafts; your strong model checks, fixes and finishes. Most of the quality at a fraction of the cost — both bills on the ledger, per task.

⚖️

Judge across models

Send one task to all your models, then let a judge model rank the answers with reasons and write the single best one. Cross-vendor consensus — the thing only a middleman can do.

All of it on your own keys. Keys, prompts and files never leave your machine; every call is billed by the provider to your account.

Multi-task × multi-model, at a glance

Your tasks down the side, your models across the top. LLManager fires them all in parallel, manages every conversation, and routes each task to the model that does it best for the least money.
tasks done 0 tokens saved 0 ≈ spend saved $0 running…

What LLManager does

Not a chat window and not a translator — a manager that handles the whole conversation and workload between you and the models.
🎛️

Manages the conversation

You write intent in plain words. LLManager shapes the right, lean instruction for each model, sends it, checks the answer held up, and keeps the thread — so you never hand-craft a prompt again.

🧵

Many sessions, one place

Dozens of live model conversations side by side, each with its own model and task. Pop any one out into its own window. Refresh, and they're all still there.

🗂️

Many tasks, many models

Run a whole batch of tasks across a whole set of models as a matrix. See who wins each cell on quality, speed and cost — then keep the best.

💸

Every token earns its keep

Light tasks go to your free models (Groq free tier, local Ollama); only the hard ones reach a paid flagship. Each task is routed to the cheapest model that still clears the quality bar, prompts are trimmed automatically, and the tally shows exactly what you saved.

🖥️

Everywhere you work

A desktop tray app (mac / win / linux) and a mobile app (iOS / Android). It reaches into what you're actually working on and runs the models right there — no copy-paste round trips.

🔐

Your keys, your data

Runs locally and talks straight to your providers on your own keys. Your prompts and content never pass through us. Bring one key or ten.

On every device you work from

One account, one workflow — wherever the work is.
Desktop · mac / win / linux
Tray cockpit + CLI
Lives in your menu bar. Hotkey to summon, drop in a task, fan it across models. Scriptable from the terminal too.
Mobile · iOS / Android
Pocket console
Speak or paste a task on the go; LLManager runs it across your models and hands back the winning answer. Share-sheet from any app.
Reaches your workspace
Runs on what you're doing
Point it at a repo, a doc, a dataset — it operates the models on that context directly, and brings the results back where you need them.

You pay for the models. We make sure it counts.

This isn't "remove please and thank you." At high spend and high volume, the money leaks in structural places — LLManager finds and fixes each one, on your real traffic.
🧊 Cache the repeated context−28%
Your fixed rubric, few-shot examples and schema are re-sent every call and re-billed as fresh input. Marked cacheable, the repeated block drops to ~10% of its cost.
🔀 Route by difficulty−24%
The easy majority of calls go to a model 10–20× cheaper that still clears your verified quality bar; the flagship only sees the hard tail.
🧠 Cap wasted reasoning−34%
On reasoning models, uncapped thinking is often the majority of the bill — tokens you pay for but never read. Held to what the task needs.
📏 Constrain output + 📦 batch−15%
Hard schema and max-tokens ceilings, and non-interactive traffic through the batch API at ~half price.
Typical stacked effect on a high-volume workload — same verified quality
up to −68% of your model bill
Illustrative levers. LLManager measures your actual number from your own traffic and only routes where verification confirms quality holds.

⬇️ Download the desktop app

Your keys, your machine — opens and closes like any app, remembers your setup, and tracks the real cost per accepted task.
⊞ Windows — installer macOS 🐧 Linux All downloads →

Unsigned for now, so first launch shows a one-time SmartScreen/Gatekeeper prompt → More info → Run anyway. Prefer source? pip install "handover[desktop] @ git+https://github.com/ravemm-hub/meerada" then meerada app.

Free core. Premium cockpit.

The measuring engine and command-line tools are open-source and free forever. LLManager is the premium cockpit on top.

Free · open-core

$0
  • The prompt engine (lean, per-model)
  • meerada ask — one model from the CLI
  • meerada fan — parallel from the CLI
  • The live Grade board & Arena
  • Visual multi-task × multi-model cockpit
  • Mobile app & workspace control
Get the free core
Premium

Meerada LLManager

Pricing soon
  • Everything in Free
  • Visual cockpit — many sessions, one place
  • Import history · switch models with it · fork
  • Attach folders · relay · cross-model judge
  • Live cost ledger + CPAT per session
  • Desktop tray + mobile apps
  • Runs models on your workspace directly
Join the private beta
We'll email you a build. No card, no spam.