Feature Requests

Artifacts currently have no storage beyond the viewer's own browser, so nothing they produce survives reliably or can be shared between users. Even a simple serial number generator becomes unusable: the counter is lost when the browser cache is cleared, and two people using it get different numbers. Artifacts need access to a database or persistent memory, shared across everyone who opens the artifact.

Summary Enable ClickUp Brain to generate executable “mini‑apps” from natural language prompts and workspace context. These apps should read and write ClickUp data (Lists, Tasks, Docs, Dashboards, Custom Fields, Tags, Comments, Attachments) with permission‑aware access and run as dynamic, interactive tools embedded in ClickUp. Support both internal workflows (inside ClickUp) and external use (lightweight client portals, prototypes, single‑task utilities). Problem Brain is strong at reasoning and content generation, but many high‑value workflows still require stitching views together, manual data transforms, or repetitive actions across Lists/Docs/Tasks. Teams need small, purpose‑built interactive tools that bind to live ClickUp data, execute multi‑step actions, and share outputs without leaving ClickUp or building separate integrations. Proposed Solution Introduce Brain‑generated mini‑apps that: Bind to data: Connect to Lists, Tasks, Docs, Dashboards, Saved Filters; expose fields and relations; support query, aggregate, and live updates. Extend to external data: Connect to Google Drive, GMail, Google Calendar, and other data sources. Provide interactive UI: Forms, tables, charts, buttons, wizards; adapt based on context (assignees, statuses, tags, custom fields). Execute actions: Create/update tasks and docs, set statuses, add comments, trigger automations, post to Chat, write to custom fields. Are permission‑aware: Enforce roles/custom roles/sharing for all reads and writes. Are deployable: Launch from Chat, Task, Doc, Dashboard, or standalone panel; share internally or externally with controlled access. Are composable: Save as templates, version apps, branch/test changes, reuse across Spaces. Example Mini‑Apps NDA Generator: Pick a template, fill variables (party names, dates, jurisdiction, term), render a final NDA, attach to the Task/Doc, post a Chat message, and open a “Sign & Track” task with due dates. Marketing Spend Planner: A calculator bound to campaign Lists and budget fields; model scenarios, generate a summary Doc with charts, and create tasks to adjust budgets and assign owners. Contract Clause Analyzer: Paste text or select a Doc; the app highlights risky clauses, suggests redlines, and opens subtasks for each recommended change, citing sections. KPI Narrative Builder: Read Dashboard widgets; explain trends, anomalies, and risks; propose follow‑up tasks with owners and dates; post a summary to the channel. Spec‑to‑Plan Converter: Pull a Doc, extract milestones, produce a dependency graph and sequenced subtasks, each citing source sections; dry‑run changes, then commit. Bug Triage Console: Intake reports (attachments, text), group by component/severity using workspace fields, mass‑create/update tasks, and publish a triage summary back to Chat. Governance & Safety Admin controls: Enable/disable Brain mini‑apps; define scopes (space/folder/list); approve data bindings and actions; set external sharing rules. Audit & observability: Log prompts, context accessed, data mutations, and app versions; exportable for compliance. Data boundaries: No cross‑workspace leakage; retention and deletion aligned with ClickUp policies; strict permission inheritance. Cost/performance: Rate limits, quotas, resource controls; “dry run” previews for bulk changes. Success Metrics Reduced time‑to‑action from prompt to committed changes. Lower app‑switching; faster decisions in Chat/Tasks/Docs. Improved data quality via structured inputs and governed bulk updates. Adoption of mini‑apps across teams; increased external collaboration.

We are building an ecosystem of ClickUp Super Agents, starting with Absencia and Onboardia, and need natural participation in ongoing Chat threads without repeated mentions. Submitted following Niko's recommendation in ClickUp Support ticket #2769821. Expected behaviour • Once an agent is @mentioned, clearly addressed by its plain-text name, or joins a conversation, it follows that thread within its permitted scope. • It receives and reads later replies, including contextual follow-ups such as “What do you think?” with no name or @mention. It then answers the latest relevant question, or acknowledges it with an appropriate native emoji or brief reply. • This also works between agents. When one agent directly addresses another, the recipient acknowledges that message. Acknowledgments are terminal; they must not create an endless agent-to-agent loop. • Replies remain in the original thread, authored by the actual native Super Agent, with genuine recipient mentions where appropriate. Following a thread must not expand permissions or grant operational authority. Requested native capability Please provide durable Super Agent thread subscriptions and a reliable “reply posted in a followed thread” event, independent of repeat mentions. Ideally this should include automatic enrollment, permitted-channel controls, follow/unfollow or mute controls, restart/catch-up behaviour, deduplication, native reply/reaction support, and bounded agent-to-agent acknowledgment. Reading every new message in a followed thread should be separate from deciding whether to interrupt or reply. How this differs from Ambient Replies ClickUp Support explained on 5 October 2026 that Ambient Replies covers follow-ups from the same person after a real @mention. The remaining gaps are replies from other thread participants, automation-started conversations, plain-name enrollment, lasting subscriptions and subscription controls, and bounded agent-to-agent acknowledgments. Support also explained that a guard may leave follow-ups unanswered and selects one agent per message. We need each relevant subscribed agent to receive the event, with a separate relevance decision and loop limits controlling its response. This request extends native conversation delivery beyond the current Ambient Replies scope. Working external workaround We currently use a small Python background worker on Render with one instance and persistent SQLite state. Within explicitly permitted channels, the ClickUp Chat API discovers roots and reads thread replies and native tagged users. Mentions, clear names and native participation establish durable per-agent subscriptions. New messages enter a deduplicated persistent outbox keyed by workspace, channel, thread, source message and agent. Polling is bounded and rate-limited, with a grace period for native handling. Where delivery is needed, the worker sends a technical envelope with source identifiers to the native agent's verified private invocation DM. The agent fetches the real thread context and replies or reacts in the original thread. Reasoning stays with the native ClickUp Super Agent; there is no external LLM. Employee narratives are not copied into the envelope or stored in the relay database/logs. Delivery intent is recorded before sending, uncertain outcomes are reconciled before retries, and native replies/reactions are independently verified. Human questions take priority, and self-messages and terminal acknowledgment loops are suppressed. We have verified unnamed follow-up replies by both native agents in our configured shared channel, restart deduplication, and bounded mutual acknowledgment. Coverage is scoped, not workspace-wide. We are retaining this relay for now and deferring migration to native handling. Why native support matters This should not require customers to operate extra hosting, credentials, polling, persistent state and deployments. A native implementation should preserve permissions, privacy and original-thread context while making event delivery observable and reliable. Acceptance should cover same-user and different-user follow-ups without re-mentioning, automation-origin threads, more than one relevant agent, restart/catch-up without duplicates, and terminal agent-to-agent acknowledgments. We would be happy to test native support so we can eventually retire the external relay.

I would like to be able to manually edit artifacts directly, without having to ask Brain to make every change. The editing experience should be similar to Canva or PowerPoint, allowing me to: Move, resize, and reposition individual elements; Edit or rewrite text directly; Delete or add elements; Change formatting, layout, and visual properties; Rearrange elements freely; Make small corrections or adjustments myself without generating a new version through Brain. The goal is to have full manual control over the artifact after it has been created, while still being able to use Brain for larger or more complex changes when needed. This would make the artifact experience much more flexible and efficient, especially for presentations, documents, diagrams, and other visual deliverables where users often need to make quick, specific adjustments themselves.

support for TypeSafe’s Jev for bounded decisions like task classification, priority scoring, and workflow routing, while keeping generative models for writing and reasoning. Let us define allowed choices, scoring rubrics, and confidence thresholds, with uncertain results escalated to a reasoning model or human review. The goal is faster, lower-cost agent workflows without maintaining a custom integration, with logged decisions and existing permissions and approval requirements preserved.

Add ClickUp AI functionality to HIPPA-compliant workspaces.

Problem More and more of our work runs through terminal agents and wrappers: Claude Code, Codex CLI, Cursor, pi and custom orchestrators. Brain is the one assistant that actually knows our company, with workspace context, installed skills, memory, connected apps and permissions. But it only exists inside the ClickUp UI and Brain MAX. So we either keep switching between tools, or we rebuild that context badly in other agents. The ClickUp MCP server doesn't solve this. It gives another model raw access to tasks and docs. It doesn't give us Brain itself: its skills, memory, Super Agents, connected-app search, or its judgment about our workspace. Request A first-party Brain CLI and agent endpoint: Auth: brain login using OAuth or a device-code flow tied to my ClickUp account. Tokens are scoped to my existing permissions, with SSO/SAML support and admin controls. Interactive mode: a brain REPL in the terminal with the same capabilities as Brain in ClickUp: skills (system and user-installed, so new skills show up automatically), memory, search across the workspace and connected apps, and write actions using the same confirmation rules. Headless mode: something like brain -p "..." --json for scripts and CI, with streaming output and structured results. Agent interoperability: expose Brain as an MCP server and/or agent-to-agent (A2A) endpoint, so Claude Code, Codex, Cursor and other harnesses can delegate work to Brain as a sub-agent. Today they can only call low-level ClickUp tools. Super Agents: invoke or chat with workspace Super Agents from the CLI. Local context bridge (opt-in): let Brain read files from the local working directory, so it can, for example, update a task with a diff summary or check a spec against the code. Governance: every CLI action shows up in the ClickUp audit log the same way UI actions do, permissions are identical to the web app, and admins can enable or disable the CLI at workspace level. Why it matters • Brain becomes the company's context layer for every agent, not just one more chat window. • Skills and memory built in ClickUp become reusable everywhere, which makes the Brain seat worth a lot more. • Without this, teams will keep piping ClickUp data into other models, and ClickUp turns into a database rather than the intelligence layer. Example workflows • In Claude Code: "ask Brain what the acceptance criteria are for this ticket and which compliance rules apply," then implement. • A CI step: brain -p "summarise this release and post it to the #releases channel" • From a terminal: brain "what's blocking the DK launch?", answered with full workspace and connected-app context.

I use the GitHub MCP a lot for projects and mockups with my agency, and every time I send a Brain message to inspect something, modify something, etc., it tries to create an artifact instead. That gets in the way of that MCP tool significantly and slows us down on working with our clients quite a bit. Would love to be able to either toggle off or set explicit rules about when an Artifact is intended to be created.

Everytime you update a dashboard AI card it creates a brand new chat, it really should just keep an ongoing running chat as my AI chats are going crazy and I keep losing chats I actually need.

We need to be able to pin/save and rename AI chats so you can easily continue a discussion or piece of work over time.