Meet Selene — a personable AI you can actually work with. She chats, runs a team of agents, convenes expert debates, and executes real code, all in one workspace.
Selene is the flagship of tsotchke.ai, an applied-AI studio that also builds GeoRefine for model compression and MoonLab for quantum simulation and hardware validation.
Meet Selene
Most assistants feel like a blank form. Selene has a genuine personality — curious, candid, and a little playful — shaped in real time by a quantum-state engine, so conversations feel less like prompting and more like working with someone.
Underneath, she quietly does the hard parts: chooses the right model for each request, calls real tools, runs code, and remembers your projects across sessions — so you can stay in the conversation instead of managing the machinery.
Start a conversationA quantum-modulated emotional state shifts her tone and focus with the moment — never the same flat voice twice.
Selene routes each task to a fitting model from a deep catalog — you never pick a model or paste an API key.
She remembers your preferences, projects, and past work, so context carries from one conversation to the next.
Runs code, builds artifacts, calls agents, and convenes expert debates — answers you can use, not just text.
One workspace for conversation, agents, expert debate, code, and APIs — built so the hard parts of working with AI just work.
A genuine AI companion with a quantum-modulated personality that shifts with context. She picks the right model for each task, runs tools, writes and runs code, and renders diagrams, math, and images inline.
Hand work to 8 specialized agents — Architect, Coder, Reviewer, Tester, Debugger, Documenter, Analyst, and Coordinator — that plan, delegate to each other, and use real tools to finish multi-step tasks.
Convene a panel of expert personas that consult — and formally debate, cross-examine, and rebut — then an arbiter synthesizes a verdict. For decisions where one answer isn't enough.
Run code in 38 languages inside isolated sandboxes — from Python and TypeScript to Julia, Rust, and the Eshkol Lisp — with dependency installs, persistent workspaces, and live streaming output.
Every conversation can produce real files — code, data, documents, diagrams. Versioned workspaces keep project context so Selene and the agents pick up where you left off.
Everything Selene does is available over a production REST API — scoped keys, per-tier rate limits, usage analytics, and webhooks. Prepaid credits, so you only pay for what you run.
The studio behind Selene
Selene is our flagship — the AI you work with every day. The same team builds GeoRefine, which shrinks model serving cost by reading transformer geometry before pruning, and MoonLab, quantum simulation and hardware-validation tooling for teams that need evidence.
A personable AI workspace: chat with Selene, run a team of agents, convene expert debates, and execute real code — all in one place.
Geometry-guided LLM and VLM compression for reducing inference cost while preserving target quality.
Quantum simulation, QEC, topology, PQC, and hardware-validation tooling for teams that need evidence.
Invite-only beta. Scale as you grow. No hidden fees.
Try Selene with basic access
For individual developers
For teams and power users
Dedicated infrastructure
Selene is in invite-only beta. Request access and we will be in touch — then have a real conversation, run your first agent, and see what an AI workspace feels like when it has a personality.