Changelog

Track shipped product updates, release notes, and major experience changes in Tulsk.

August 28, 2026 — Free Starting Balance Is Now $2

The Free plan’s one-time agent-work balance is $2, down from $5.

Why: the larger balance was being farmed with throwaway sign-ups rather than used to try the product.
Unchanged: Free is still free forever — unlimited projects and tasks, no credit card — and the balance is still one-time, not monthly. Spend it and agent work stops until you upgrade; the rest of the workspace keeps working.
Top-ups still start at $5, and paid plans are untouched: Starter $50/month, Pro $120/month.

August 17, 2026 — Landing Pages in Every Language

The marketing site now speaks all seven supported languages, not just English.

Seven Sections Translated: Always-on agents, channels, hiring an agent, the workspace computer, memory, skills, and connections are now translated into French, Spanish, Japanese, Korean, Simplified Chinese, and Traditional Chinese.
Localized Navigation: Sign-in, sign-up, and “Open workspace” links across every marketing page now follow your language.
Use Cases in Your Language: The use-case gallery headings and workflow links are translated too.

August 15, 2026 — Agent Skills, Templates & One Simpler Agent Page

Everything about configuring an agent now lives on one page — and the Skills tab finally shows the skills that actually run.

Core Pillars

The Skills tab lists what the agent really loads at run time, and opens each file in an editor with a folder tree so you can change it in place.

Skills That Actually Run

Pick a built-in role when you create an agent — including an SEO role that ships with a working audit skill.

Start From a Template

A rich editor for agent instructions, stored as markdown. Click anywhere in the box to start typing.

Instructions, Properly Edited

Per-agent monthly budgets are gone. Agent work draws on your single workspace allowance, the same as chat.

One Allowance, No Per-Agent Caps

⚡

Agent Builder retired: The separate builder surface is gone. There’s no engine picker and no agent-type picker — one configuration page, one generic agent model.

Other Improvements

Chat Sidebar: Pinned chats get their own section, per-chat actions moved into a proper menu, and rename/pin/delete are now reachable on touch.
Thread Details: Threads show when they started; the noisy per-message usage chip is gone.
Settings Grouping: Billing and Usage now sit together under Workspace.
Workspace Memory: Rendered as readable sentences instead of raw JSON.

August 8, 2026 — Connections, Grants & Inline Approvals

Agents can reach your tools — but only the ones you grant, and only with your sign-off on writes.

Core Pillars

Connections are owned by the workspace and granted per agent. An agent can only use what you’ve explicitly given it.

Per-Agent Connection Grants

Resolve an approval inline in chat instead of detouring to a settings page.

Approvals in the Conversation

When an agent needs a connection it doesn’t have, it files a request instead of failing. Approve it and the work resumes where it left off.

Denial Becomes a Request

Bring your own MCP servers and HTTP APIs. They’re granted by name exactly like managed apps, with headers encrypted at rest.

Custom Connections

⚡

Waiting doesn’t mean stopped: An external write that needs approval is queued, not blocked. The agent gets a queued result and keeps working on everything else.

Other Improvements

Cost in Chat: Per-reply usage chips and thread totals now show real dollar cost, and survive a page reload.
Agent Picker in the Composer: Switch which agent a thread is bound to mid-conversation.
Integration-Event Triggers: Start an agent run from an event on a connected app.
Connection Status: Composio connection states are mirrored onto the integrations list so you can see what’s actually live.

August 2, 2026 — Every Workspace Gets a Computer

Your agents now work on a real machine with a durable filesystem — and so does chat.

Core Pillars

A single persistent machine per workspace. Files stay where your agents left them, readable whether anything is running or not.

One Computer Per Workspace

A dock in chat shows the workspace filesystem and any running preview, without leaving the conversation.

Files & Preview, Always There

If a dev server stops, a Start button brings it back by replaying how it was started. Restarting it never costs an agent turn.

Previews That Restart Themselves

Workspace Chat can run commands, read and write files, and expose ports itself — no agent needed in the loop.

Chat Drives It Directly

⚡

Free plan clarified: Free includes $5 of agent work once per account — not $5 every month. Starter and Pro refresh their allowance each cycle.

Other Improvements

Wider Model Access: Platform models now resolve through a single provider layer, so new models become available faster.
Honest Metering: Compute and storage draw down your plan allowance, sampled so you’re never billed for time nobody observed.
Personal Memory Controls: Decide what Tulsk remembers about you, separately from workspace knowledge.

July 30, 2026 — Live Previews & Faster Agent Runs

Watch what an agent builds while it’s building it.

Core Pillars

When an agent exposes a port, a preview appears in chat and on the task — docked, expanded, or collapsed.

See What Your Agent Built

Preview URLs are authenticated, so work in progress isn’t publicly reachable.

Previews Stay Private

Machines stop on their own when nothing is happening, and you’re metered on running time only.

Sleeps When Idle

Runs go through a single execution path instead of a separate dispatch hop, cutting startup latency.

Faster Agent Runs

Other Improvements

Simpler Notifications: Nudges now come through the app inbox and email directly.
Faster Workspace Load: Layout data is fetched in parallel, the board streams in, and routes show skeletons instead of blank space.
Accessibility Pass: A round of keyboard, focus, and interaction fixes across the app.

June 4, 2026 — Agent Channels: Slack, Telegram & WhatsApp

Talk to your agents where your team already talks.

Core Pillars

Chat with your agents over Slack, Telegram, and WhatsApp.

Three Platforms

A channel conversation reaches the same agents, skills, and workspace memory as the app.

Same Agents, Same Memory

Bind an agent to an alias per connection and mention it in a Slack channel to route work to it.

Mention to Address

Per-platform connect instructions, with a dedicated Slack app separate from your workspace integration.

Guided Setup

⚡

Readable replies: Long answers are chunked and rendered as native markdown per platform, and delivery failures surface instead of disappearing.

May 27, 2026 — Bring Your Own LLM Keys

Use your own model credentials, and get a better chat editor while you’re at it.

Core Pillars

Connect Anthropic, OpenAI, or Gemini credentials and run agents on your own account.

Your Keys, Your Models

Added and promoted to recommended for fast, cheap runs.

Gemini 3.5 Flash

Reasoning-model output is folded into a collapsible “Thinking…” block instead of flooding the thread.

Thinking, Collapsed

Type / in agent chat for quick actions, with a stop button that keeps whatever was already written.

Slash Commands

Other Improvements

Images in Task Descriptions: Paste screenshots straight in, then resize them in place.
Edit and Delete Comments: Fix a typo or remove a comment after posting.
Enter to Send: Enter posts a comment, Shift+Enter adds a newline.
Blog: A /blog section with launch posts and proper SEO metadata.
Mobile Chat: A floating button and full-screen chat sheet on mobile web.

April 17, 2026 — Global Search, File Attachments & Inbox

A faster way to find work, share context, and stay on top of what’s changed.

Core Pillars

Search every task in your workspace from one place. Debounced, paginated, keyboard-friendly.

Workspace Search

Drop images, PDFs, CSVs, and docs into any comment. Images render inline; files become tidy cards with size and download.

File Attachments

The old Activities page, reimagined. A single feed for task updates, mentions, and agent replies — with one-click mark-all-read.

Inbox

Follow any task to get notified on updates. Subscriber avatars appear in the task header so ownership is obvious.

Task Subscribers

⚡

Workspace Chat Schedule from Chat: Hit the new calendar button in the expanded Workspace Chat chat to turn any conversation into a recurring report — no detour to settings.

April 14, 2026 — V3 Billing & Credit Wallet

A complete rebuild of billing around how teams actually use Tulsk: cluster-based plans with a monthly allowance and top-up credits for the months you go big.

Core Pillars

Pay per cluster, not per seat. One price, up to 6 agents per cluster.

Cluster-Based Pricing

See exactly how much of your plan you’ve used this month with color-coded alerts as you approach the limit.

Monthly Allowance Bar

Burst past your allowance without getting blocked. Top up credits, see how many tasks they’ll buy, and roll over unused balance.

Overtime Credits

Owners and admins manage billing. Members see their workspace context and plan status without access to payment actions.

Role-Gated Controls

⚡

Payment Health at a Glance: If a payment fails, you’ll see an amber banner and a one-click path to the Stripe portal. PAST_DUE and UNPAID states are surfaced instead of silently breaking.

Other Improvements

Usage Dashboard: New /settings/usage page with token consumption timeseries, top-run breakdowns, and per-period filtering.
Token Estimation on Agent Runs: Agent comments now show estimated token cost so you can spot expensive runs before they add up.
Billing Reconciliation: Daily cron heals any DB/Stripe drift automatically. On-demand admin reconciliation endpoint for support cases.
Clean Cluster Delete: Removing a cluster now syncs the billing side atomically — no more orphaned Stripe items.
Checkout Session Recovery: If the success page loads before the webhook fires, the UI now reconciles the session client-side so you never see a stale “unpaid” state.

April 10, 2026 — Structured Skills & Agent Health

Give your agents real skills, and know exactly when they’re healthy.

Core Pillars

Define skills with forms and specs — name, inputs, outputs, and acceptance criteria. Agents pick them up automatically.

Structured Skills

Workspace Chat’s harness can now research and evolve skills based on how they perform in real tasks.

Workspace Chat Autoresearch

Not sure if your cluster is live? One-click health check on the cluster page. Heartbeat polling reduced to save compute.

Manual Health Check

Pick from 20+ common zones (UTC, LA, London, Tokyo, Sydney, …) and Workspace Chat schedules land at the right local time.

Timezone-Aware Schedules

Other Improvements

Settings Kanban for User: Personal settings reorganized into a card-based layout so appearance, profile, and security are easier to scan.
Members in Workspace Settings: Member management moved into Settings → Members for consistency with other workspace admin tasks.
Japanese Locale Fix: Renamed the locale from jp to ja to match ISO 639-1. Existing links redirect.
Agent Builder Localization: Agent Builder pages and share flows now localized across all 7 supported languages.

April 3, 2026 — Workspace Chat Scheduled Reports

Stop writing status updates. Tell Workspace Chat what you care about once, and she’ll send you the right report to Slack or email — every morning, every Monday, or whenever you want.

Core Pillars

Pick “every weekday at 9 AM” or write your own schedule. Workspace Chat handles timezone math so reports land when your team actually reads them.

Set It and Forget It

Scope reports to specific projects, statuses, priorities, or individual tasks. No more sifting through noise.

Only What Matters

Reports show up in your Slack channel, your inbox, or both. Formatted natively for each — not a wall of plain text.

Delivered Where You Work

Hit preview, read the report, adjust if needed. Hit “Send Now” when you want it right away instead of waiting for the next run.

Preview Before It Ships

⚡

Works out of the box: Workspace Chat ships with a Daily Briefing and a Weekly Report already set up. Toggle them on — that’s it. You’ll get your first report tomorrow morning.

Your Reports, Your Rules

Plain English Prompts: Tell Workspace Chat “summarize what agents shipped this week and flag anything blocked” — she figures out the rest.
Unlimited Schedules: Daily standups, weekly exec summaries, Friday retros — create as many as you need.
One-Click Pause: Going on vacation? Pause any schedule and pick it back up when you’re back. Nothing lost.

Focus on the Right Signal

Pick Your Projects: Report on one project, three projects, or everything in the workspace.
Filter by Status & Priority: Only care about high-priority in-progress work? Set it once.
Drill Into Specific Tasks: Skip the filters entirely — just select the exact tasks you’re tracking.

Slack & Gmail, Native Formatting

Slack: Reports land in your channel formatted with proper Slack markup — bold, links, sections. Long reports are trimmed with a link to the full version.
Gmail: Clean HTML emails that look professional. Subject lines auto-generated. Multiple recipients each get their own copy.
Both at Once: Same report, two channels. Your IC team gets Slack, your investors get email.

Never Miss a Beat

Run History: Every report execution is logged — when it ran, how long it took, whether delivery succeeded.
Failure Alerts: If a delivery fails, you’ll see exactly what went wrong and which channel had the issue.
No Duplicates: Even if the system fires twice, you’ll only ever get one report. We handle the dedup.

April 3, 2026 — Workspace Chat: AI Project Manager

Workspace Chat goes from a chat assistant to a full AI project manager that handles the complete project lifecycle — from first idea to project closure.

Core Pillars

Say “Start a new project” and Workspace Chat creates the project, generates tasks, sets up a sprint, budget, stakeholders, quality plan, and identifies risks — all in one conversation.

Start-to-Finish Orchestration

Workspace Chat remembers your preferences, decisions, and project context across every conversation. She never starts from scratch.

Persistent Memory

Workspace Chat watches your projects around the clock — retrying failed agents, flagging stuck tasks, detecting scope drift, and alerting on health drops.

Automated Monitoring

New Workspace Chat dashboard with health score trends, sprint velocity charts, agent performance bars, and recent activity timeline.

Visual Dashboard

⚡

One Message, Full Project: “Start a new project called Fitness Tracker” → Workspace Chat creates 27 tasks across 4 phases, sets up Sprint 1, allocates budget, identifies 3 risks, and asks “Want me to assign agents?”

New Capabilities

Smart Agent Routing: Workspace Chat picks the best OpenClaw agent based on skills, track record, and availability.
Auto-Retry on Failure: If an agent fails, Workspace Chat retries automatically — adjusting instructions, trying a different agent, or escalating to you.
Sprint Management: Create sprints, track scope changes, detect drift, and measure velocity.
Task Dependencies: Set task ordering so blockers are tracked automatically.
Risk Register: Add risks with probability, impact, and mitigation strategies. Workspace Chat monitors them.
Project Health Score: See how your project is doing at a glance with trend tracking over time.
Change Control: Mid-sprint scope changes go through impact analysis before approval.
Decision Tracking: Workspace Chat captures decisions in real-time and posts them to the relevant task.
Sprint Retrospectives: Auto-generated retros with action items created as tasks.
Project Closure: Full summary with metrics, lessons learned, and recommendations.
Task Breakdown: Break any project into phases and tasks with dependencies automatically.
Project Kickoff: Structured brief with success criteria, initial risks, and agent recommendations.

Enterprise PM

Stakeholder Register: Track people, roles, influence levels, and communication preferences.
Project Budget: Per-project budget with category breakdown and spend tracking.
Quality Management: Acceptance criteria, defect tracking by severity, and threshold alerts.
Change Control: Change request → impact analysis → approve, reject, or defer.

Workspace Chat Dashboard and Schedules

Workspace Chat Dashboard: Visual overview with health trends, sprint velocity, agent performance, and activity timeline.
Workspace Chat Schedules: Manage automated daily briefings, weekly reports, and custom schedules with pause/resume.
Quick Action Buttons: Sprint Status, Quick Briefing, and Risks on the left of chat. Dashboard and Schedules on the right.

Workspace Chat Chat UX

Auto-Introduction: Workspace Chat auto-opens and introduces herself when you haven’t talked to her before.
Health Indicator: Color dot on Workspace Chat’s avatar showing project health at a glance.
Proactive Persona: Workspace Chat leads with insight, not data. Makes recommendations, flags problems, references past conversations.

March 30, 2026 — Agent Builder

Design your OpenClaw agent team visually.

Visual Builder: Drag-and-drop canvas to configure agent personas, skills, and roles.
3D Characters: See your agents come to life with 3D character previews.
Share Builds: Share your agent configurations via public links for collaboration.
Cluster-Aware: Build up to 6 agents per cluster, matching your deployment capacity.

March 14, 2026 — Agent Memory, Direct Chat & Smart Scheduling

Agents now remember what they’ve worked on, can be chatted with directly, and schedules run on time.

Core Pillars

Agents remember prior conversations on each task. Follow-up @mentions build on previous context instead of starting fresh.

Agent Memory

Chat with any agent directly from their profile page — no task required. Conversations persist across sessions.

Agent Direct Chat

Schedules now fire at the exact time. Edit schedules inline with confirmation before saving.

Smart Scheduling

Usage charts, response time trends, one-click health checks, and model testing.

Agent Dashboard

⚡

Agent Memory: @mention an agent on a task, then follow up later — the agent remembers what it already did. No more repeating yourself.

Other Improvements

Run History by Task: Agent runs grouped by task with clickable links and collapsible details.
Health Check: Verify your agent is online with a single click from the overview page.
Model Test: Send a test prompt to confirm your agent is responding before assigning real work.
Analytics Charts: Usage and response time trends on the Costs page.
Task References: When agents mention “Task #18” in comments, it becomes a clickable link.
Workspace Chat Chat: Delete confirmation, improved chat history, updated avatar.
Faster Page Loads: Optimized data loading across agent and task pages.

March 4, 2026 — Multi-Agent Clusters & Workspace Chat as PM

Run multiple agents on a single cluster and let Workspace Chat coordinate them as your AI Project Manager.

Core Pillars

Run up to 6 agents on a single cluster, each with their own identity and workspace.

Agent Clusters

Workspace Chat now acts as your AI Project Manager with awareness of what every agent is working on.

Workspace Chat as AI PM

Set up automated reports in plain English — “send me a summary every Monday at 9am.”

Scheduled Reports

Agents can delegate work to each other. Workspace Chat traces the full chain of who did what.

Multi-Agent Delegation

⚡

Faster Streaming: Agent responses now appear significantly faster in real-time.

Other Improvements

Agent Monitoring: Health charts and compact grouping in the agent list.
Agent Lifecycle: Activate or deactivate agents in any cluster with a single toggle.
Multi-Language: Updated pages with 7-language translations.
Refreshed Documentation: Modernized layout with interactive components and improved visual hierarchy.

February 3, 2026 — Autonomous Agent Workflows

The transition to autonomous agent workflows for complex execution and cross-team project collaboration.

Autonomous Workflows: Agents now browse, search, and analyze independently until a task is complete.
Multi-Language Support: Support for international campaigns and localized content.
Collaboration Overhaul: Improved task threads and real-time syncing across workspace members.

August 22, 2025 — Initial Release

The initial release of Tulsk — foundational AI team coordination and workspace infrastructure.