A Local-First Agent for Private and Cost-Effective Knowledge Work
Perplexity, Tuesday, August 25th, 2026
Perplexity Research describes co-designing a harness and post-trained model that run entirely on the user's device.
Perplexity Research describes the design of Portable Computer, a local-first agent in which the entire stack runs locally by default.
The model, harness, conversation and trajectory all live on the user's machine rather than in a service. Work that genuinely needs the outside world, such as web search, connectors, or escalation to a stronger advisor model in the cloud, is invoked only when necessary and always gated by the user.
Sensitive data therefore never leaves the device without permission, and local models carry no inference fee, making the system private and cost-effective by construction. The post covers what an effective local-first agent requires from both the harness and the post-trained model.