engblogs

Ten years in, this post explores how Ink & Switch channels AI, open-source software, and interactive art—embodied by Tenfold—to push public good while navigating climate, the reproducibility crisis, and the health of journalism and democracy.

Introducing OpenAI Privacy Filter: an open-weight, on-device, bidirectional token-classification model for fast, context-aware PII detection and redaction across long text to enable privacy-preserving AI workflows.

A technical overview of workspace agents in ChatGPT, powered by Codex, enabling shareable, memory-enabled automation that orchestrates complex workflows across tools like ChatGPT and Slack with enterprise-grade controls.

Demonstrates how WebSockets unlock a persistent, end-to-end agent loop in the Responses API, slashing overhead and boosting speed to over 1,000 tokens per second through in-memory caching, reduced network hops, and asynchronous tool execution.

Introducing OpenAI Privacy Filter: an open-weight, on-device, context-aware PII detection and redaction model designed for fast, privacy-preserving text processing with configurable deployment options.

OpenAI Privacy Filter is a compact, on-device, bidirectional token-classification model that detects and redacts PII in long, unstructured text with context-aware accuracy, fast single-pass processing, and configurable privacy controls for deployment across diverse workflows.

Codex-powered workspace agents in ChatGPT transform team knowledge and repeatable workflows into shared, memory-enabled automations across tools like ChatGPT and Slack, with enterprise governance and controls.

A practical guide to designing and deploying ChatGPT workspace agents that automate repeatable workflows using triggers, processes, tools, and guardrails.

Introducing workspace agents in ChatGPT: Codex-powered, cloud-based shared agents that automate complex workflows, integrate across tools like ChatGPT and Slack, and enable enterprise governance and memory for teams.

How WebSockets standardizes a persistent transport for the Responses API to accelerate Codex's agent loops, enabling in-memory caching, reduced network hops, and asynchronous tool execution to boost token throughput from about 65 TPS to over 1,000 TPS in production.

ChatGPT for Clinicians provides free access for verified U.S. clinicians to an AI-powered clinical assistant that streamlines documentation, research, and care workflows with evidence-based search, HIPAA-ready options, and HealthBench Professional benchmarking to improve safety and patient outcomes.

Shows how WebSockets in the Codex Responses API enable a persistent connection with in-memory state caching to dramatically cut agentic workflow latency and push inference throughput beyond 1,000 tokens per second.