What's New in Deelo: Monthly Product Updates (May 2026)

Davaughn White

Most of what shipped in Deelo this month is not the kind of thing you write a flashy launch post about. It is the AI Assistant remembering your chat session after a reload. It is a billing surface that tells you exactly which features are blocked when you run out of credits, rather than failing silently. It is Vibe generating with a streaming intent indicator so you know what the model is actually doing while you wait.

The theme of May was reliability and clarity, not net-new surface area. Below is the actual changelog of what landed in the last four weeks, what is moving next, and where to look in the product if you want to feel the differences.

Reliability and Security

Two changes that should not be visible if they are working. They are working.

AI Assistant chat session persistence fix. Chat sessions in the AI Assistant now reliably survive page reloads, browser restarts, and tab switches. The previous behavior dropped active session context in a small percentage of cases — particularly around long-running tool calls — which forced users to repeat themselves. Sessions are now hydrated from the server on mount with a single source of truth.
IDOR hardening on chat session endpoints. A round of authorization checks closed a class of insecure direct object reference (IDOR) issues on chat session routes. Every read and write to a session now verifies the requesting user owns the session, scoped to their team. There is no observable behavior change for legitimate users; the change matters for security posture and audit.
2FA gate on sensitive account actions. A two-factor authentication gate is now enforced on a defined set of sensitive account operations — primarily around session revocation and access changes — to align with the same protections already in place on billing and team-management endpoints.

AI Assistant

Generation tools available from the Assistant. The AI Assistant can now invoke the same generation tools available inside Vibe, which means you can ask the Assistant to draft, refine, or extend content without switching apps. The tools execute in the Assistant's tool-call layer and stream their output back into the conversation, so a long generation no longer feels like the chat is frozen.
Credits-exhausted action button. When a credit-gated tool call cannot run because the team is out of AI credits, the Assistant now surfaces a specific, in-line action button to top up — instead of returning a generic error. The Assistant continues the conversation, the user understands why the tool did not run, and the path to resolve is one click.

Vibe

Streaming intent indicator. Vibe now shows a streaming "intent" line above the generation while the model is working — a short phrase describing what step the model is currently doing ("reading project files," "writing component," "applying changes"). The signal is small, but the difference in perceived responsiveness on a 30-second generation is large: you stop wondering whether something is stuck.
Two-bubble chat UX. The Vibe chat now uses a two-bubble pattern — a top bubble for the model's plan and intent, and a bottom bubble for the actual streaming output. The split makes it easier to scan a long response and to find where the model started writing code versus where it was still reasoning about the change.
Per-project UI state reset on openProject. A small but high-impact change: opening a different Vibe project now correctly resets the per-project UI state — open files, active tab, scroll position — instead of leaking state from the previous project. This was a recurring source of confusion when switching between two projects in quick succession.
Image uploads in chat. You can now upload images directly into a Vibe chat. The image is attached to the message and available to the model as context for the next generation — useful for design references, screenshots of bugs, or marking up an existing UI.

Billing and Credits

Credits-exhausted banner across the platform. A persistent in-app banner now appears across the workspace when a team is out of AI credits, with a direct link to top up. Previously, the indication only appeared inside the apps that consumed credits, which meant a user could spend several minutes troubleshooting why an AI feature "was not working" before realizing the balance was zero.
Inline credits-exhausted CTAs on nine surfaces. Beyond the global banner, nine specific UI surfaces — including the AI Assistant chat input, Vibe's generate button, content drafting modals in the Marketing app, and the AI-suggested actions in the CRM — now surface a credit-aware inline CTA when the team balance is exhausted. The intent: catch the user at the moment of action with an explanation and a top-up button, not after the action silently fails.
402 credit-blocked handling, consistently. Every frontend app that calls an AI-credit-consuming endpoint now handles the 402 response with a toast that includes the specific `blockedReason` from the server. "Out of credits" is now distinguishable from "this feature is not on your plan" — same status code, different actions.

Under the Hood

Two changes worth flagging even though they are not user-facing:

npm lifecycle hook cleanup. The build pipeline no longer chains local-only cleanup scripts (port-freeing, zombie-killing) into `prebuild`. Those scripts ran in CI and killed legitimate CI children, occasionally flaking deploys. They are now manual `npm run` commands and the CI pipeline is steadier.
Onboarding email and assistant session work continued. The multi-channel onboarding flow — welcome email, AI Assistant kickoff session, checklists, and auto-open of the first app — saw refinements across the four touchpoints. New signups are completing the first-week checklist at a higher rate; we will share the specifics once the comparison window is long enough to be honest about.

What's Next

A short, honest look at what is in flight — not what we hope to ship someday, just what is in progress this sprint and likely to land in the next monthly update:

Redis-backed rate limiter (TASK-2741). The current rate limiter is in-memory and resets on deploy, which makes burst protection inconsistent across instances. Moving it to Redis is the right shape for a multi-instance deployment and is in progress.
Next sprint themes. Continued AI Assistant tool expansion, additional Vibe project-scoped settings, and a round of CRM workflow polish based on the last quarter's customer feedback. We will cover specifics in next month's update once they are actually shipped.

Found a Bug, Have a Request?

If something in this month's update is not behaving the way it should, or you want to ask whether a specific behavior is a bug or a feature, the fastest path is the in-app feedback button in the bottom-right of any Deelo workspace. Every report is read, and the ones that catch a regression usually ship a fix in the same sprint they are reported.

Frequently Asked Questions

Does the AI Assistant chat session persistence fix require any user action?: No. The fix is server-side: sessions are now hydrated from the server on mount with a single source of truth, so any active chat in the Assistant survives page reloads, browser restarts, and tab switches automatically. If you experienced lost session context before, the next time you open the Assistant should feel different — particularly for long-running tool calls.
What does the Vibe streaming intent indicator look like?: While Vibe generates, a small intent line appears above the streaming output — a short phrase describing what the model is currently doing ("reading project files," "writing component," "applying changes"). On long generations (15-30 seconds), this signal dramatically reduces the perception that something is stuck. The bottom bubble continues to show the actual streaming output.
How do credits-exhausted CTAs work across the platform?: When your team's AI credit balance hits zero, a persistent workspace banner appears at the top with a top-up link. Beyond that, nine specific UI surfaces — Assistant chat input, Vibe generate button, Marketing drafting modals, AI-suggested CRM actions, and others — show credit-aware inline CTAs at the moment of action. The intent: explain why the action did not run and provide a one-click resolution.
Will I notice the IDOR hardening change?: No, intentionally. The change closes a class of authorization vulnerabilities on chat session endpoints — every read and write now verifies the requesting user owns the session, scoped to their team. There is no observable behavior change for legitimate users. The fix matters for security posture and audit, not for daily experience.
When does the Redis-backed rate limiter ship?: It is in progress this sprint and likely to land in next month's update. The current in-memory rate limiter resets on deploy, which makes burst protection inconsistent across instances — moving it to Redis is the correct shape for a multi-instance deployment. We will share the launch confirmation in the June 2026 changelog once it is shipped.
Can I upload images into Vibe chats now?: Yes. Image uploads landed this month. Attach an image to a Vibe message and it becomes available to the model as context for the next generation — useful for design references, screenshots of bugs you want fixed, or marking up an existing UI. Drag-and-drop into the chat input works, as does the upload button.

Try the new Vibe and AI Assistant

Open Vibe to see the streaming intent indicator and two-bubble chat UX in action, or try the AI Assistant with generation tools and the new credits-aware CTAs. Both are included in every Deelo workspace at $19/seat/month. Sign in and explore — or start a free trial if you are not on Deelo yet.

Start Free — No Credit Card

Related pages

Explore More

All software categories Compare alternatives Best software by industry