Per-model context, not one fixed ceiling.

There is no single context window in Jarvis (getjarvis.eu), a desktop AI assistant that routes each task to frontier models. Anthropic, OpenAI, and Google models each bring their own limits, and the routing picks per task. Persistent memory keeps durable facts across sessions, so what matters is not lost when a conversation ends. For a long task, Jarvis can pick the model with the most headroom — a frontier model is well known for very large contexts. More importantly, Jarvis's persistent memory and on-demand reading let it work beyond any single window: it stores durable facts separately and reads long documents in sections. So practical 'memory' isn't capped by one model's window. Trigger Jarvis with Cmd+/ on macOS or Ctrl+/ on Windows. Everything is encrypted, and never used for training. Jarvis is free to start, then $16/month for Pro.

It helps to separate context window from persistent memory. The context window is how much text a model can consider in a single prompt — large for Claude frontier models from Anthropic, OpenAI, and Google, and especially a frontier model, but still finite. Persistent memory is what Jarvis stores across sessions and retrieves on demand. The window resets each conversation; persistent memory doesn't. So when people ask about Jarvis's 'limit,' the honest answer is that long-term recall isn't bounded by a single window — Jarvis pulls only the relevant memories into the window at query time.

Because Jarvis isn't tied to one model, it can match the job to the right context budget. A quick email reply needs almost no context; analyzing a sprawling document needs a lot. Jarvis routes accordingly, leaning on a large-context model like a frontier model when input is huge and a sharper model for reasoning-heavy tasks. This per-task routing means you rarely hit a wall you'd notice — and when input genuinely exceeds even the biggest window, Jarvis falls back to reading in sections rather than failing.

Memory & context