This week we sharpened three priorities: make AI agents safer and immediately useful with real integrations, smooth how your team reads and acts on work content, and tighten the publishing and notification experience so nothing slips through the cracks. The result: agents that use central credentials, a live memory UI, inline document previews inside tasks, and a healthier newsletter and send pipeline.
Below are the changes you’ll notice when you sign in this week.
Agents, credentials, and memory — run AI with real integrations
Agents can now pick credentials directly from the central MCP app catalog instead of relying on ad-hoc per-organization config. New integrations (Gmail, Drive, Asana, Slack, Notion, GitHub) are available to agents the moment they’re added to the catalog; you no longer have to recreate connector entries in every workspace.
We made OAuth more flexible: a single Google / Asana / Slack OAuth connection can back credentials across multiple organizations, so power users don’t need to reconnect the same account in each org. The MCP runtime is smarter about the caller environment, so widgets and metadata render correctly whether the client is hosted by Claude (mcp-ui) or OpenAI.
The upgraded agent memory system is live in production. You get the new memory UI (Table and Graph views), with tier badges, segment coloring, and trust scores exposed in the main site. That makes it easier to inspect what your agents remember and why, and to iterate on memory design.
Why this matters: agents now have immediate, governed access to real apps and a clearer memory surface to operate with, which reduces setup time and lowers the risk of unpredictable behavior.
Smarter task and document workflows — inline previews, subtasks, and Google Tasks sync
Pasting a Google Doc or public Notion page into a task description now renders an inline, read-only preview panel right in the task drawer. You can read the brief without opening a new tab, and the document text is fetched and cached on the task as searchable metadata so agents and site search work with the content directly.
From the preview you can:
Create a Playbook pre-populated with the document’s content and a traceable link back to the source URL.
Generate actionable subtasks from the document (AI-extracted list, up to 20 items, default 8) and attach them as children of the task in one click.
Refresh the preview manually to re-sync the source content.
Supported sources: public/shared Google Docs and public Notion pages. Notion workspace URLs that block embedding fall back to an Open document button. Google Docs links that point at a specific tab are honoured — the preview and any generated Playbook or subtasks use the tab’s content when available.
We also shipped two-way Google Tasks sync. Each user can opt in from Settings → Calendar, pick a default Google Tasks list (or create one inline), and have OKiDO tasks assigned to them appear in Google Tasks and Google Calendar. Changes flow both directions: edits in OKiDO push immediately; changes in Google pull every five minutes; completing a task on either side marks it done on the other. The sync is per-user and opt-in, so teammates aren’t affected unless they enable it.
Other workflow improvements include task due-date reminders and a per-member notification preferences panel (opt in/out per notification type: task assigned, task due, mentions, comments). These help keep work visible without noisy inboxes.
We also shipped a set of reliability and UX fixes that reduce friction and surface issues sooner:
Sprint-view task assignments update the card immediately without requiring a refresh.
The audio recorder now supports language detection beyond English-first transcription.
MCP widgets render correctly in Claude-hosted sessions and general MCP client compatibility has been improved.
Google Tasks OAuth callbacks and mirror sync were made robust: the OAuth callback no longer fails with ORG_ID_REQUIRED, and mirrored tasks now show readable plaintext notes (with a link back to the OKiDO task) instead of raw TipTap JSON. Inbound edits from Google are converted back into TipTap format and the appended URL is stripped so routine polls don’t look like user edits.
Plus, we tightened build- and deploy-time checks so missing dependency issues are caught earlier and rare deployment/build failures are less likely to impact you. Those platform reliability investments should make updates smoother and reduce surprise errors during upgrades.
What to try this week
Give an agent a connected credential from the MCP app catalog and watch it act across connected apps without per-org reconfiguration.
Paste a Google Doc into a task description, use the preview to generate a Playbook or subtasks, and try assigning the generated subtasks in a sprint.
Opt in to Google Tasks sync from Settings → Calendar and see your OKiDO tasks appear in Google Calendar alongside your schedule.
If you run into anything unexpected or want help wiring an agent to your org’s MCP catalog, reach out through support or your customer success contact. We’ll keep iterating on reliability, integrations, and the parts of the product that help your team actually get work done.
