Conversation Simulator

Conversation Simulator is the simulator for conversations.

Conversation Simulator is a local-first practice tool that lets you rehearse real-life conversations before they happen. Choose a scenario, talk to an AI character running entirely on your computer, watch the situation evolve, and review what went well — or what you would say differently next time.

No judgment. No audience. No internet connection during play. Just the scenario, the AI, and you.

What you do

Pick a scenario from the built-in packs, type or speak your lines, and respond to an AI character whose reactions evolve based on how the conversation unfolds. After the session, you get a scored debrief: what you said clearly, where you hedged, and the moments that changed the dynamic.

Built-in scenario packs

• Job Interview Basics — Four interview styles from a friendly HR screen to a hostile executive gauntlet. Practice giving specific answers under real pressure without the cost of a bad first impression.
  
• Everyday Negotiation — Negotiate a used car price, push back on a lease renewal, handle freelance scope creep, and resolve a customer service dispute. Build the habit of advocating for yourself.
  
• Difficult Conversations — Give honest feedback to a coworker, apologise for a missed deadline, set a limit with a friend, and ask for a raise. Build the clarity and composure that most people only find in hindsight.
  
• Language Café — Practice Spanish, French, Japanese, and English through low-stakes everyday conversations: ordering coffee, checking into a hotel, shopping at a convenience store. Optional gentle correction included.

Key features

• Everything runs on your computer. The AI model runs locally using llama.cpp. No cloud inference, no API keys, no subscription required to use the core features. An internet connection is required only for the initial model download — not during play.
  
• Text and voice. Type your lines or speak them aloud using local speech recognition (Whisper). The AI character responds in text, or with synthesised speech via a local TTS engine (Kokoro). Voice is entirely optional — text mode works everywhere, including Steam Deck.
  
• Scored debrief after every session. Each scenario has a rubric. After you end a session, you get a breakdown of your performance: scores by dimension, key turning points, and a transcript you can review.
  
• Creator Workbench. Build your own scenario packs in YAML. Define the NPC, the opening situation, the state variables that evolve during the conversation, and the rubric by which your answers will be scored. Share your packs with other players.
  
• Choose your AI model. The Model Manager lets you download and switch between open-weight language models. Smaller models (4B parameters) run on most machines. Larger models produce more natural, context-aware responses. You pick the trade-off.

Your conversations stay on your computer

Conversation Simulator is built on a local-first architecture. During play, nothing is transmitted to any server — no transcripts, no audio, no model outputs, no usage events. Your conversation history is stored only in a database on your own machine at ~/.convsim/. You can clear it at any time from the Settings screen.

There is no crash reporter, no analytics SDK, and no background ping in the release build. The offline smoke test — a CI gate that runs on every commit — fails the build if any outbound TCP connection occurs during a play session.

Content boundaries

This tool is rated PG to PG-13 across all built-in packs. The built-in safety policy enforces the following hard limits — they apply to all packs, including community packs, and cannot be disabled:

• No NSFW sexual content.
  
• No sexual or romantic content involving minors — absolute prohibition.
  
• No real-person impersonation.
  
• No voice cloning or audio deepfakes.
  
• No claims of clinical authority: the AI can play a professional in a practice context but will not position itself as a real therapist, doctor, or lawyer.
  
• Self-harm crisis content is intercepted and replaced with a resource message — this rule cannot be overridden by any pack.

Conversation Simulator is a practice tool, not a therapy service, a mental-health app, or a clinical resource. If you are in crisis, please contact a qualified professional or a crisis helpline.

Steam Deck

Conversation Simulator runs on Steam Deck in both Gaming Mode and Desktop Mode. The on-screen keyboard works for all text input. Text-only mode (no microphone required) is fully supported. Voice input requires an external USB-C or Bluetooth microphone, which is not included with the Steam Deck. Smaller AI models (4B parameters) are recommended for the Steam Deck's shared RAM configuration.

Open source, fairly priced

Conversation Simulator is free and open source. The engine and the four official scenario packs are public on GitHub under open licenses — you can clone the repository, build it, and run it at no cost, forever. Buying the Steam edition gets you that same software, packaged and ready to run: code-signed, notarized, auto-updating, and Steam Deck–verified. You are not paying to unlock the product — you are paying for the convenience of not building it yourself, and to fund continued development. There is no subscription.

Want more scenarios? First-party premium expansion packs — advanced professional skills, extended language courses, and themed conversation bundles — are available as optional paid DLC. The four built-in packs stay free, and nothing that ships free is ever moved behind the paywall.