The technology behind our products, explained without the fog — what each thing is, how it actually works, and what to look for. Every page starts with a straight answer.
An AI computer operator is an AI agent that operates a computer on your behalf: you give it an instruction in plain language — spoken or typed — and it works through the steps itself, clicking, typing, and moving between applications, instead of just telling you what to do.
A credential firewall is a security layer between an AI agent and your real secrets. The agent only ever sees dummy keys; a local, deny-by-default proxy swaps in the real credential just before the request leaves — and only for destinations you have allow-listed — while logging every request for audit.
On-device speech to text is voice-to-text that runs entirely on your own hardware: a speech model on your CPU or GPU converts microphone audio into text locally, so the audio is never uploaded to a server. After the model is set up, transcription itself can run without a network connection.
An MCP server is a program that exposes tools and data to AI assistants through the Model Context Protocol (MCP), an open standard for connecting AI applications to external capabilities. The assistant calls the server's tools — search this, fetch that, analyze this — and uses the results in its answer or its work.
A code graph is a structured representation of a codebase where functions, types, and modules are nodes, and relationships — calls, imports, dependencies, effects — are edges. Instead of reading files one at a time, a tool or AI agent queries the graph: who calls this, what does it touch, what breaks if it changes.
Yes — language models can run on an ordinary CPU with no GPU at all. The keys are quantization (storing weights in ~4 bits instead of 16, shrinking memory several-fold) and CPU-native code paths such as hand-written AVX2 kernels. The result is private, local inference on everyday laptops — at smaller model sizes and lower speeds than datacenter GPUs.
Live speech translation listens to speech in one language and produces text or speech in another as the conversation happens — near-realtime, not after the fact. Modern implementations pair speech recognition with translation models; the meaningful differences are latency, accuracy, and where your audio is processed.
Prompt injection is an attack where instructions are hidden inside content an AI system reads — a web page, an email, a file — so the model treats the attacker's text as commands. Because language models process instructions and data in the same stream, no fully reliable fix exists; practical defense means limiting what a confused agent can do.
A coding agent is an AI system that works on software the way a developer does: it reads the codebase, plans a change, edits files, runs builds and tests, reads the failures, and iterates until the task is done. Unlike autocomplete, it operates in a loop with tools — deciding the next action, executing it, and learning from the result.
Speech-to-text software — dictation software — turns your speech into text and types it into whatever application you are using. What separates products: where the audio is processed (cloud or your device), what happens to your words afterwards (a rule-based clean-up or a generative rewrite), whether use is limited, which platforms are covered, and whether you eventually own it.
Voice-to-text — also called speech-to-text — converts spoken words into written text. It is used two ways: live dictation, where words appear as you speak them in whatever application you are in, and transcription, where recorded audio becomes a text record. What separates the tools: where the audio is processed, whether use is limited, and which platforms are covered.
A lightweight PDF reader does the reading job — open, scroll, zoom, search, copy — without the editing suite, cloud sync, ads and account of a suite. On Android the difference is measurable: install size, ads or purchases on the listing, whether you can read without signing in, and whether the app holds the Internet permission.
Context saving is keeping a coding agent's context window for the actual change by serving structural answers — definitions, callers, dependencies, blast radius — from a persistent code graph instead of having the agent re-read source files. Across 955 jobs on 18 public repositories, a typical job took 2,221 tokens in one call, not 13,321 in two.