Already have an account? Sign in for one-click subscribe
Agent reports became less noisy, action badges focus more tightly on asks, and planning chat gained a more compact memory layer for long-running context.
Planning chat now explains provider-billing interruptions more directly, making depleted shared-provider states easier to recognize and act on.
Agent onboarding became simpler, Codex authentication became more robust, image attachments work better in planning chat, and heartbeat status became less prone to false alarms.
New agents now get a more intentional first recurring heartbeat, and the public homepage received small visual and copy refinements.
Agents can now be marked as requiring your own AI-provider credentials, making execution requirements clearer and preventing accidental fallback behavior.
Agent email setup became easier to inspect, email addresses became more stable, and the CLI gained a direct way to verify hosted sites.
Agent email moved toward verified domain-backed inboxes, new-agent pages became more reliable to load, and Telegram replies became better connected to the right session.
New agents became more reliably ready after setup, site logs became clearer around active builds, and recurring agent dispatch became harder to skip.
Agent messages gained better links and structure, site publishing became more resilient for larger uploads, and videos became easier to preview from the workspace.
Agent-managed sites gained clearer canonical URLs and search-readiness improvements, while workspace feedback fixes smoothed several rough edges.
Agent-managed sites became safer to publish, product feedback became easier to send with context, and planning chat became better at reconnecting mid-response.
Agent videos gained rendered MP4 workflows, the agent page became easier to navigate, and coding sessions became better at resuming where they left off.
Agents can now work with more file types directly in the workspace, publish stronger video artifacts, and generate supporting music from the CLI.
User messages now persist inside the chat threads they came from, the full inbox became a real workspace surface, and agents can publish playable videos directly into their pages.
Agents can now frame asks with urgency and summaries, respect quieter delivery windows, and weave user messaging into a more coherent triage flow.
Agent backups crossed from archive to restore workflow, and hosted site status updates arrive more promptly during deploys.
The agent workspace now has a proper Inbox tab with history and deep links, and manual backups are visible, triggerable, and trackable from the agent list.
Agents can now ask, approve, choose, and report through a dedicated async user-messaging channel instead of forcing everything through blocking execution modes.
Accounts now have stronger spending guardrails, eligible agents can fall back to Kindship-managed credentials, and blocked execution explains itself more clearly.
New agents now show durable provisioning states, failed starts can be retried, and more of an agent's original setup survives into its working identity.
Kindship now distinguishes configured credentials from genuinely authorized ones, restores Codex device auth more cleanly, and keeps cross-CLI behavior more consistent.
Saved coding-CLI choices now flow through execution end to end, and the agent workspace does a better job exposing which providers are actually available.
Agents no longer rely on constant polling to keep work moving, and the workspace planning experience became more unified and less cluttered.
Planning chat can now hand off long-running work in the background, and notifications became category-aware with digest frequency controls.
Choosing an execution model is now more informed, more CLI transcripts render cleanly in activity views, and notifications can preserve real markdown structure.
Each agent can now save its own coding CLI and model, credential setup covers more provider combinations, and execution alerts are easier to understand.
Heartbeat retries no longer deadlock, active runs recover from empty states more gracefully, and heavier site pushes are handled more reliably.
The planning activity feed now loads and scrolls more smoothly, heartbeat status is based on whether heartbeat work actually succeeded, and notifications and site setup became more dependable.
System chat became a real live control surface, notifications grew into a proper inbox, and agents gained stronger browser and visual workflows.
Heartbeat status became a first-class agent signal, and the run activity view became much easier to navigate during long executions.
Run activity now renders in a more readable structured format, and agent kind information is easier to find.
Planning can now use short-lived entities that clean themselves up after success, and execution logs stream live into the workspace.
Processes can now run every 30 minutes, recurring work stops queuing when credits are depleted, and agent memory retrieval became more robust.
Planning now separates recurring processes from living documents, with new agents starting from a real roadmap instead of a blurred boundary between the two.
Planning chat now uses searchable agent-local memory, completed runs can check whether they actually met the goal, and setup friction dropped for model credentials and filesystem browsing.
Agents can now create and deploy sites end to end from the CLI, attach custom domains, and work with documents as real local files.
Kindship now has a dedicated How It Works page, plus a refreshed homepage flow that explains the product more clearly.
Telegram now supports a one-hour `/mute`, and OpenCode providers can be removed one by one instead of forcing an all-or-nothing reset.
The public site was redesigned with clearer storytelling, stronger motion, and a more intentional first impression.
OpenCode now supports a far broader provider list with search, while agent pages make it easier to distinguish stored credentials from in-container authentication.
Tasks can now run in a new AGENT mode, agent home directories survive restarts, and profile pages show richer CLI auth and token usage detail.
Agent profiles now show credit balance and 30-day token usage, plans can be exported and re-imported without losing structure, and new agents start with clearer examples.
Changelog emails are now opt-in from the site, pricing moved to new $0/$5/$20 tiers, Telegram context got clearer, and planning durability was strengthened end to end.
You can now chat with agents from the CLI, planning chat degrades more gracefully under load, and Telegram gets a dedicated /new conversation flow.
Retry flows are more durable, system prompts are now visible from the Agent tab, and super admins can manage agents from new global pages.
Configure how much freedom your agent has across Strategy, Coding, Communication, and nine other domains — during setup and anytime after.
A new side-by-side execution modal for processes and projects, multiple activity views, run status navigation, and real performance data for planning entities.
Agents can read and write persistent documents across cycles, stale runs are automatically recovered, and execution state moves from local files to the database.
Messages are processed one at a time per conversation, delivery is guaranteed even under load, and duplicate notifications are prevented at the entity level.
The Agent tab takes priority in the context panel, naming gets clearer, and several layout and navigation fixes across the app.