OpenChamber.dev

OpenChamber ✨ I started this project because I fell in love with @opencode as a TUI coding agent, but realized I personally didn’t enjoy living in the terminal all day. I wanted easier remote access, something that worked nicely from a phone/tablet, and a UI that felt more natural to use day-to-day. What started as a fan-made UI slowly grew into something much bigger. OpenCode now has its own web UI and desktop app, but I still keep building OpenChamber because the project naturally grew into its own thing over time. I like experimenting with workflows, tooling, and UX ideas around agents - and people seem to enjoy using it too 🙂 github.com/btriapitsyn/openc…

Great example of generative UI at incredible speed!

Keeping you in the loop. 5 OpenChamber releases in 11 days: - on a limit, auto-switch to another account or model - 160+ integrations, one-click MCP install - agents draw charts and mockups in chat - Gitea, Forgejo, Gerrit, stacked PRs - E2E-encrypted push with agent replies - @ a session for context

We read three years of agent memory research and kept what was actually measured. Haiku 5.5 went through 15,603 papers and picked the 3,283 about agent memory. Complex memory systems often do no better than keeping the full history and searching it well. What works: - Keep everything and put the effort into search. How you search moved accuracy by 20 points. How you save, by 3 to 8. - Mark facts that changed. Plain search returned an outdated fact 15 to 40% of the time. With a clear “this replaces that” rule, almost never. - Hand memory to the agent yourself. One agent had everything it needed in memory and looked it up 0 times in 114 turns. - Don’t let rules get summarized away. Claude Code’s /compact on Sonnet 4.6 kept 53% of safety rules after one pass and 10% after five. - Save where each memory came from. In one test that blocked every attack. Instructions in the prompt cut attacks by 3.2 points at most. In another test, plain RAG was as accurate as Mem0 and cost 8.4 times less. Two things are still broken. An instruction planted in Claude Code’s memory fired in later sessions 81.7 percent of the time. Asked to forget something, none of six systems did it more than 1.5% of the time. Full breakdown: openchamber.dev/blog/agent-m…

Congrats to the @useblacksmith team! OpenChamber's CI runs on Blacksmith, and our Mac builds now take half the time: the desktop app went from 10 minutes to 5, iOS from 7 to 3.5. CI used to be the slow part of shipping a release. Now it's me.

Someone asked for a quieter update pace. No! v2.2.0 is out, with tons of feature requests implemented. Seatbelts are in Settings.

If you missed, here are most of the things announced at the @OpenAI #DevDay 2026 – Dots – always-on agents that learn what matters to you and pick up tasks on their own (for example, triaging bugs or answering emails) – GPT-6.1 Sol – almost Astra-level, but 5x cheaper per token. Available now. – Ultrafast – a premium speed tier: up to 8x in Codex (300 tokens/sec) and up to 6x in the API. GPT-6 Astra only for now. – Pro $500 – a new plan with 25x the limits of Plus and access to Ultrafast, for 500 bucks. – Agents API with computer use – plus multi-agent from Codex, tool search and context compaction. – Decisions API – the model answers predefined questions with fixed answer options - a.k.a. Jev from @typesafeai – Codex – cloud runs from any device, voice control in the CLI, code review in the desktop app, and Codex Security Cloud. – Plugins: ChatGPT is opening up its internal platform: sidebar panels, file viewers, automations via MCP Events – Office – Space for teams, Pages as shared docs for humans and agents, collaborative slides with export to PowerPoint/Google Slides, @ChatGPT in Slack and Teams, a meeting-recording plugin. – Sign in with ChatGPT – your subscription works at 16 partners (Devin, Notion, Vercel, etc.), with a limit for each one. – Marketplace – companies can now spend their already-committed OpenAI budget on software from 32 partners: Adobe, Figma, Salesforce, Harvey, CrowdStrike, etc.

Here is a сollection of 3,348 videos made with Opus 5.5 Motion graphics, explainers, product promos, 3D scenes, music videos, and even games, all made with Opus 5.5 and posted on X since last week 605 of them come with the author's prompt, from the source post or from the replies where they shared it Every video plays as the original X post, and the prompt copies in one click (at least you can get decent results on your own videos) You know what to do with it 🤗 data.openchamber.dev/opus-5-…

I expected the OpenCode 2 migration to be a bug hunt. It was mostly us adopting our own code. The few things we did hit, @opencode team fixed right away. @thdxr offered to help before we even asked. Thank you ❤️

OpenChamber 2.0 is out, on OpenCode v2. Skills, agents, MCP servers and plugins now apply the moment you save them. No restarts. My restart button is unemployed. Plus Code Mode, web search cards and a stats page. openchamber.dev/blog/opencod…

#meme #Jev 😄