Smarter starts, faster feedback
23 March – 9 April 2026
Platform version: v0.6.54
Themed Prompt Categories
When you start a new AutoPilot conversation, you now see themed prompt categories — Learn, Create, Automate, and Organize. Each theme button opens a popover with 5 contextual prompts tailored to the category. Personalized prompts are distributed across themes automatically. ↗
Live Timer Stats
A live elapsed timer now appears in the AutoPilot thinking indicator while the AI is processing — showing "23s", "1m 5s" and so on after a 20-second threshold so quick responses aren't cluttered. Once the response completes, a "Thought for Xm Ys" badge freezes in place below the message. The duration stays with the conversation, so you can always see how long a response took — even after refreshing the page or revisiting an older chat. ↗
Redesigned Onboarding
The onboarding wizard has been completely redesigned with an Autopilot-first flow. A polished 4-step experience — Welcome → Role selection → Pain points → Preparing workspace — collects your role and pain points to personalize Autopilot suggestions from the start. Pain points are reordered based on your selected role (e.g. Sales sees "Finding leads" first), and steps sync with URL params so browser back/forward just works. ↗
Copy Your Prompts
You can now copy your own prompt messages in AutoPilot with a single click. A copy button appears on hover — right-aligned below the message — using the same CopyButton component already available on assistant responses. No more manual text selection when you want to reuse or share a prompt you've written. ↗
✨ Improvements
Spend credits to reset AutoPilot daily rate limit — Hit your daily message cap? Spend credits to keep going without waiting for the reset. (#12526)
Rate-limit tiering system — AutoPilot now supports configurable rate-limit tiers so different user plans get different message caps. (#12581)
Extended thinking execution mode — OrchestratorBlock gains an extended thinking mode for deeper, multi-step reasoning. (#12512)
Automatic AgentMail setup — AgentMail now works out of the box. Your email address is created automatically the first time you need it — no API keys or manual setup required. (#12537)
Session-level dry-run flag — Autopilot sessions can now be flagged as dry-run at the session level. (#12582)
Generic ask_question AutoPilot tool — A new tool lets AutoPilot ask structured clarifying questions mid-conversation. (#12647)
Git committer identity from GitHub profile — E2B sandbox commits now use the user's real GitHub name and email. (#12650)
Create → dry-run → fix agent generation loop — Agent generation now follows an iterative create, dry-run, and fix cycle for better results. (#12578)
All 12 Z.ai GLM models via OpenRouter — Access every Z.ai GLM model through OpenRouter integration. (#12672)
include_graph option for find_library_agent — Fetch the full graph structure when searching library agents for debugging or editing. (#12622)
Cursor-based message pagination — AutoPilot messages load newest-first with cursor-based pagination for faster load times. (#12328)
React integration testing — Vitest + React Testing Library + MSW test infrastructure added. (#12667)
Cost tracking for system credentials — Platform now tracks costs incurred by system-managed credentials. (#12696)
🎨 UI/UX Improvements
Auto-reconnect after device sleep — AutoPilot chat reconnects automatically after your device wakes from sleep. (#12519)
Marketplace card descriptions — Cards now show 3-line descriptions with a fallback color when no image exists. (#12557)
Hide placeholder during voice recording — Placeholder text hides when AutoPilot voice recording is active. (#12534)
Array field layout fix — Fixed array field item layout and added FormRenderer Storybook stories. (#12532)
Realistic AutoPilot suggestions — Replaced unrealistic AutoPilot suggestion prompts with practical ones. (#12564)
Show all agent outputs — Agent results now display all outputs, not just the last one. (#12504)
Notification follow-ups — AutoPilot notifications gain branding, UX improvements, persistence, and cross-tab sync. (#12428)
Horizontal scroll for JSON output — Builder JSON output data now scrolls horizontally instead of overflowing. (#12638)
Onboarding polish — Refined AutoPilot onboarding with branding, auto-advance, soft cap, and visual polish. (#12686)
🐛 Bug Fixes
Prevent graph execution stuck — Fixed stuck graph executions and steered SDK away from bash_exec. (#12548)
Sink input validation — AgentValidator now validates sink inputs to catch wiring errors early. (#12514)
Docker Node.js upgrade — Upgraded from EOL Node.js v21 to v22 LTS in Docker images. (#12561)
Thinking block preservation — Thinking blocks are now preserved during transcript compaction. (#12574)
AIConditionBlock error handling — Errors are now raised instead of silently swallowed. (#12593)
Host-scoped credentials — Fixed credential resolution for authenticated web requests. (#12579)
Duplicate tool name disambiguation — OrchestratorBlock now disambiguates duplicate tool names. (#12555)
Gmail recipient validation — Gmail blocks validate email recipients before making API calls. (#12546)
Tool call circuit breakers — Added circuit breakers and intermediate persistence in AutoPilot. (#12604)
Prompt-too-long retry — Automatic compaction, model-aware compression, and truncated tool call recovery. (#12625)
Credential loading fix — Fixed AutoPilot credential loading across event loops. (#12628)
Duplicate block execution prevention — Prevented duplicate execution from pre-launch argument mismatch. (#12632)
Tool output file reading — Fixed tool output file reading between E2B and host. (#12646)
Dry-run mode propagation — Dry-run mode now propagates to special blocks with LLM-powered simulation. (#12575)
Double-submit prevention — Prevented duplicate side effects from double-submit and stale-cache races. (#12660)
Silent DataError fix — Wrapped PlatformCostLog metadata in SafeJson to fix silent DataError. (#12713)
Last updated
Was this helpful?