Explore what's new in AI

Chronicle gives Codex passive memory over what you have been doing

The feature turns prior computer activity into usable context, which unlocks more surprising workflows than plain chat history. Persistent memory is becoming a real product primitive.

Gemini adds file generation and export in chat

Gemini can now create files directly from a prompt and export them in the requested format. That makes it much more useful for docs, spreadsheets, and quick handoff work.

Google rolls out Preferred Sources for Search in all supported languages

Users can now star outlets and steer Top Stories toward the sites they trust more often. That changes search result shaping without touching ranking itself.

ComfyUI adds Qwen Edit workflows for multi-angle character generation

Upload one image and the workflow can generate multiple views automatically, either locally or on Comfy Cloud. That is a handy production path for controlled character iteration.

Cabinet ships out of beta as an open-source knowledge base and AI workbench

Cabinet bundles knowledge base, task automation, code fixes, and planning into one BYO-AI app. The CLI-first launch makes it easy to slot into existing workflows.

vitest-evals rebuilds around a harness-first pi-ai testing flow

The PR turns evals into normal-looking tests that run once, then judge the recorded session. That makes agent regression testing more inspectable and less flaky.

OpenCode adds session undo so bad prompts can be rolled back

A single revert command can back out a bad turn and restore the prior state, which is a small but high-leverage safety rail for agent sessions.

Cline rewrites its core as an SDK with pluggable providers and CLI

The team rebuilt Cline around a new SDK, then layered the CLI and extension on top. That should make agent profiles, model providers, and plugin support much easier to evolve.

Hermes Agent v0.12.0 ships The Curator Release and a smoother update path

NousResearch pushed a fresh release with an update command and a new Curator label. It is a small but concrete sign the local agent stack is still moving fast.

Warp breaks down its block model for terminal rendering and command handling

The deep dive explains how Warp turns terminal output into blocks, character grids, and sumtrees. It is a useful reference for anyone building an agentic terminal UI.

Jakub Krehel ships /oklch-skill for palette generation and contrast checks

The skill handles OKLCH conversion, migration, palette generation, and accessibility checks. Handy if you want to work in perceptual color space instead of hex guesswork.

Cursor explains how it keeps its agent harness faster, smarter, and cheaper

Cursor treats the harness as a product: dynamic context, model-specific tuning, live A/B tests, and guardrails measured against keep rate. Good blueprint for agent infra teams.

Zed 1.0 adds bookmarks and command-palette commit diff viewing

Bookmarks replace breakpoint hacks, and you can jump to any commit diff from the palette with HEAD, a SHA, a branch, or a tag. Small upgrades that save real editor friction.

Hermes Agent runs natively in LM Studio and loads local models on demand

The integration auto-discovers local models, picks the right context size, and adjusts reasoning level per model. That makes local agent workflows less manual.

Linear adds Releases to track deploys with version, environment, and status

Teams can now manage software releases inside Linear and give agents deployment context without leaving the issue. It bridges planning with the actual rollout state.