A substantial fortnight. Your credentials now live in an encrypted vault that agents can use without ever seeing them, the sidebar has been rebuilt around organising your work into folders, and Player One watches its own disk, processor and memory so a heavy task cannot bring the machine down. Anthropic's newest models are available, Claude Opus 5 among them, and underneath there are hundreds of fixes: chats open faster, long conversations cost less, and your agents communicate far more efficiently.
New
Watch a browser session, or take the wheel
When an agent is working in a browser you can open it and watch on demand, step in to sign in yourself, then hand control back so it carries on. Files it downloads on the remote side can be brought back to you.
New
The sidebar, rebuilt: folders and subfolders
Create folders and nest folders inside them, then drag chats in or move them straight from a chat's own menu. Pinned items can be reordered, a folder can be kept open at the top where you always see it, and long folders load in pages so a busy sidebar stays quick. Your layout is saved to your account rather than to a single browser, so it follows you between devices, and your agents can create folders and file chats away when you ask.
New
Select many chats and act on them at once
Tick several chats and move or delete them together instead of one at a time. Bulk clear-outs show their progress, carry on running if you minimise them, and can be cancelled partway. Deleting a chat removes what it spawned along with it.
New
Move chats between projects
Chats can be reassigned to a different project, in bulk if needed, so work that started in the wrong place is easily filed correctly. Folders live within their project, and progress is shown across all of them.
Improvement
Faster teamwork between your agents
Agent-to-agent messaging runs through a single dependable route, so agents reach each other and report back to you quickly, without the duplicate messages and needless sign-off replies that used to clutter a thread. A message survives a restart instead of being lost.
Improvement
Coordinators understand projects
A coordinator running a multi-step job stays within the right project and checks it can reach the files it needs before starting. Jobs begin cleanly without double confirmations, the chat that asked for the work is woken when it finishes, and a completed job is clearly flagged as successful.
Improvement
A steadier side-by-side preview
The preview pane beside your chat is more dependable: it no longer reloads while an agent is mid-reply, it picks up file changes properly instead of showing a stale copy, PowerPoint decks render, saved files keep their real names, and a browser preview closes itself when the session ends. Large tool results are shown in their own panel rather than swamping the conversation.
New
Player One now watches its own resources
Player One keeps a much closer eye on disk space, processor load and memory. When any of them run high it eases off and tells you rather than pressing on, and it will not begin a task your machine cannot finish. In short, your agents can no longer bring the box down by taking on too much.
New
A real Trash for deleted chats
Deleting a chat now sends it to a Trash you can review, filtered by project, with recursive deletion for a conversation and everything it spawned. It can finally remove broken chats that previously refused to go.
New
Memory rebuilt into the platform
Player One's memory now runs inside the platform rather than alongside it, with a dependable overnight consolidation process and sharper recall. You can read the transcript of a memory run to see exactly what was recorded, and knowledge can be scoped to a single project.
New
New models across every provider
Anthropic's Claude Opus 5 is now available to choose in any chat, alongside Sonnet 5 and Fable 5. Kimi and Moonshot became first-class providers, bringing Kimi K3 and the K2.7 coding models, and Google's Gemini 3.6 Flash and 3.5 Flash Lite joined the catalogue with OpenAI's GPT-5.5 and 5.6. Anthropic replies are also cached for a full hour, so a top-tier model costs less than it used to on a long conversation.
New
A proper Google Workspace connection
Connecting Google is now a first-class part of Player One. An agent walks you through choosing the services you want and reaches any Google API, not a fixed handful, asking for additional access as new needs come up rather than demanding everything upfront. Gmail, Drive, Sheets, Calendar and the rest are all in reach, Drive folder syncing reuses the same connection, and access is brokered so an agent uses your account without ever holding the token itself.
New
Share a credential with one project
Stored keys and passwords can be scoped to a chosen project instead of being available everywhere. Requests that cross a boundary wait for your approval, and anything pending sits at the top of the vault so it is not missed.
New
Read and produce big documents
Player One can work through a long document, spreadsheet, presentation or PDF in sections and summarise it without flooding the conversation. Agents can also produce polished PDFs and PowerPoint decks, and read, edit and merge existing files.
New
A warning before you send a key
If a message looks like it contains an API key, Player One warns you before it is sent, so a secret does not end up in a chat by accident. Gateway-owned files are also protected from agent edits.
New
Guided project setup
New projects arrive with their own workspace and a short guided setup, and it is now clear which resources belong to a project and which are shared across the instance.
Improvement
Chats open noticeably faster
Opening a conversation does far less work: the composer appears before the transcript finishes loading, and repeated background requests have been merged. Moving around the sidebar feels quicker throughout.
Improvement
Lower cost on long conversations
The instructions sent with every reply now stay byte-for-byte stable when an agent picks up a new tool, so the cache is no longer discarded mid-conversation. Long threads keep their savings instead of losing them partway through.
Improvement
A clearer model list, and new providers
The model list is paged with a full view and better search, provider accounts show a proper display name, and Kimi and Moonshot joined the curated catalogue. Thinking effort can be set per model, and any model your routing cannot reach is marked as manual only rather than failing quietly.
Improvement
Agents pick up tools without interrupting themselves
An agent that needs a tool it has not loaded now gets it mid-task and carries straight on, rather than stopping and starting again. Fewer stalls on complex requests.
Improvement
Archive a chat instead of deleting it
Conversations can be archived and hidden from the main list while remaining available, and replies that arrived from another channel are now clearly marked.
Improvement
Hundreds of fixes to the core system
Beyond the headline features this period included over three hundred individual fixes: error messages that say what actually went wrong, steadier provider handling, cleaner notifications, chats that resume after a restart, and a long list of smaller rough edges across the interface.