O que construímos depois.
Proponha ideias, vote nas que você precisa e acompanhe o status. As mais votadas sobem na lista.
Robots: Pause a run for a human decision and resume from the app
UiPath Action Center lets a robot hand a task to a person and continue after they respond (https://docs.uipath.com/action-center/automation-cloud/latest/user-guide/about-action-center). Robots run start to finish with no way to wait for approval, for example before submitting a form. First scope: an Ask for approval step that stops the run in a waiting status, notifies the workspace, and continues or fails when someone answers in the run view.
Robots: Show a run timeline with the duration of each step
UiPath Orchestrator shows execution logs per job with timestamps for each activity (https://docs.uipath.com/orchestrator/automation-cloud/latest/user-guide/logs). Our run detail lists steps and status (components/robots/RunsPanel.tsx) but not how long each took. A team tuning a slow robot cannot see which step eats the time budget. First scope: store started and finished times per step in the run log and show the duration next to each step.
Robots: Retry a group of steps together with a limit and delay
UiPath's Retry Scope reruns a whole sequence until a condition passes (https://docs.uipath.com/activities/other/latest/ui-automation/activities-index). Our retries apply to one step at a time (retries and retryDelayMs in lib/robots-schema.ts), so a flow like open menu, click, wait for result cannot be retried as a unit. First scope: a group step with up to 5 attempts that reruns its children from the first one and logs each attempt.
Robots: Add a Hover Text step that finds text and hovers over it
UiPath has a Hover activity and a Hover Text activity that searches a string inside an element and hovers it (https://docs.uipath.com/activities/other/latest/ui-automation/hover-text). Our Click step takes a selector or visible text, but nothing hovers (lib/robots-interpreter.ts). Menus that open on hover block small teams from automating many admin panels. First scope: a hover step with selector or visible text, plus a short wait for the menu to render.
Robots: Add a Toggle option to the Check or uncheck step
UiPath's Check/Uncheck activity offers Check, Uncheck and Toggle (switch the current state) (https://docs.uipath.com/activities/other/latest/ui-automation/activities-index). Our Check or uncheck step (lib/robots-schema.ts) only sets a final state of checked or unchecked. Toggle matters for switches that a site renders as a custom button where the current state is not readable. First scope: a third state value that clicks once and logs the result.
- 0
Robots: Add a Check or uncheck step for checkboxes and radio buttons
UiPath has a dedicated Check/Uncheck activity that sets a checkbox to a known state instead of toggling it (https://docs.uipath.com/activities/other/latest/ui-automation/). Our Click step toggles, so a robot that runs on a page where the box starts checked flips it the wrong way. First scope: a check step with a checked/unchecked choice that uses the page's real state, and a recorder mapping from checkbox clicks to this step.
Robots: Dismiss cookie banners and dialogs automatically before each step
UiPath handles unexpected popups with Pop-up handlers triggered during execution (https://docs.uipath.com/activities/other/latest/ui-automation/). Robots here fail or click the wrong element when a consent banner covers the page, and the user must record a manual click step for every site. For a small team that scrapes many sites this is the most common flaky step. First scope: a robot setting that clicks known consent buttons and auto-accepts JavaScript alert dialogs, with a note in the run log when it fired.
Robots: Switch between tabs and handle popups opened by a click
UiPath's Attach Browser and Use Browser scopes let a workflow move between windows and tabs (https://docs.uipath.com/activities/other/latest/ui-automation/). Our interpreter runs against a single page (lib/robots-interpreter.ts), so a link with target=_blank or a login popup leaves the robot on the old page. This blocks a lot of real portals. First scope: a `switch_tab` step (newest, by URL contains, back to first) and an option on Click to follow the new tab.
Robots: Add an Assert step that fails the run when a value or element is wrong
UiPath ships Verify Expression and Check App State activities to validate a page before continuing (https://docs.uipath.com/activities/other/latest/ui-automation/). We have If exists, which can fail on a missing element, but nothing that checks a value such as 'row count is greater than 0' or 'text contains Welcome'. Robots that silently return empty tables look like successes in the run history. First scope: an assert step with element exists, text contains and variable comparisons, and a clear error in the run log.
Robots: Add a Set variable step with text transforms (trim, regex, replace, number, date)
UiPath's Assign and string activities let a workflow reshape values between steps (https://docs.uipath.com/activities/other/latest/ui-automation/). Today our robots can only write variables through Extract or Evaluate with custom JavaScript (lib/robots-interpreter.ts), so cleaning a price or building a URL needs code. A small team wants '$1,299.00' to become 1299 without writing a script. First scope: a set_var step with a template and a list of transforms, usable in later steps as {{vars.name}}.
Robots: Add hover, double click and right click actions to the Click step
UiPath's Click activity exposes click type (single, double) and mouse button, and hover is a separate Hover activity (https://docs.uipath.com/activities/other/latest/ui-automation/). Our Click step in lib/robots-schema.ts only does a single left click, so menus that open on hover and context menus cannot be automated. Small teams hit this on dashboards with hover menus. First scope: a `mode` option (click, double, right, hover) on the click step, recorded by the recorder when it sees dblclick or contextmenu events.
Robots: Expose robots as tools over MCP
Apify runs a hosted MCP server at mcp.apify.com that lets Claude, Cursor or ChatGPT search, run and read Actors (https://docs.apify.com/integrations/mcp). Robots can already be run by Agents inside zerosoft and through /api/v1/robots, but external AI clients cannot use them. First scope: an MCP endpoint that lists the workspace's robots as tools with their inputs as the schema, runs them through the existing queue and returns the output, behind an API key with the robots:run scope. Priority: low. Effort: S. Compared with Apify.
Robots: Allow longer runs and more pages for catalog crawls
Browse AI allows tasks up to 60 minutes on every paid plan (https://www.browse.ai/pricing); Apify runs are bounded only by the plan's compute. Robots caps a run at 180 seconds, 20 pages and 1,000 list rows (lib/robots-schema.ts), so a 30-page catalog or a detail-page crawl over 100 items cannot finish in one run. First scope: a long-run mode (up to 15 minutes, 200 pages, 10,000 rows) available to scheduled and API runs, with progress saved per page so a timeout keeps what was extracted. Priority: medium. Effort: M. Compared with Browse AI / Apify.
- 0
Robots: Paginate with Load more buttons and infinite scroll in Extract list
Browse AI paginates numbered pages, 'load more' buttons and infinite scroll (https://www.browse.ai/features). Extract list and Extract table only follow a Next button (lib/robots-schema.ts pagination), so feeds and modern catalogs need a hand-built loop of scroll and click steps that does not dedupe rows. First scope: a pagination mode select (Next, Load more, Scroll) where Load more clicks until the button is gone and Scroll scrolls until no new cards appear, both deduping rows by card text. Priority: medium. Effort: S. Compared with Browse AI.
- 0
Robots: Bulk run from a CSV upload in the UI and raise the batch limit
Browse AI bulk runs take thousands of inputs uploaded from a sheet or CSV (https://www.browse.ai/docs/api/v2). Robots bulk runs exist only in the v1 API and cap at 100 inputs per batch (MAX_BULK_INPUTS in lib/robots-schema.ts). Running a robot over a list of 1,000 product URLs is the common case. First scope: a 'Run for many' dialog that accepts a CSV, maps columns to inputs, shows batch progress and lets you download all results as one CSV, with the batch limit raised to 1,000 and runs queued. Priority: medium. Effort: S. Compared with Browse AI.
- 0
Robots: Keep an accumulated table of rows across runs with a dedupe key
Browse AI stores every run's rows in a per-robot Table with tabs, filters and export (https://help.browse.ai/en/articles/10477537-how-to-view-manage-and-export-your-scraped-data); Apify appends to named datasets (https://docs.apify.com/platform/storage/dataset). In Robots each run keeps its own output and CSV, so a price history or a growing lead list has to be assembled by hand. First scope: a Data tab that merges a list variable across runs by a chosen key column, with first seen, last seen and changed columns, filters and CSV export. Priority: medium. Effort: M. Compared with Browse AI / Apify.
- 0
Robots: Email a highlighted diff when a monitored robot's output changes
Browse AI monitors send a color-coded email with added, changed and removed rows (https://help.browse.ai/en/articles/10524268-how-to-set-up-monitoring-notifications). Robots already computes the diff and emits robot.output.changed (lib/robots-diff.ts), but nobody hears about it unless they build a flow on that event. First scope: an 'Email me when the data changes' checkbox next to the schedule that sends the diff (before and after per key, rows added and removed) to the robot owner, throttled to one email per run. Priority: medium. Effort: S. Compared with Browse AI.
- 0
Robots: Send extracted rows to Google Sheets or Airtable from the robot
Browse AI keeps Google Sheets and Airtable in live sync with a robot's table (https://www.browse.ai/pricing). In Robots the only path is building a flow from the 'Robot data to Google Sheets' template, which needs a service account and only appends. First scope: a Destination section in the robot's Advanced tab that picks a Sheets credential and range and appends (or replaces) the rows of a chosen list variable after each successful run, reusing the Flow Sheets client. Priority: medium. Effort: S. Compared with Browse AI.
Robots: Route runs through residential proxies by country and handle captchas
Browse AI uses rotating residential proxies with location targeting and solves supported captchas and Cloudflare challenges (https://help.browse.ai/en/articles/10452902-what-websites-work-with-browse-ai); Apify sells datacenter, residential and SERP proxies with country targeting (https://docs.apify.com/proxy). Robots runs from the runner's own IP (lib/robots-runner.ts), so marketplaces and real estate portals block it or show the wrong country. First scope: a proxy credential type (URL, user, password) and a per-robot proxy and country setting passed to Chromium launch, then a captcha-solver credential used when a challenge is detected. Priority: high. Effort: M. Compared with Browse AI / Apify.
- 0
Robots: Generate a robot from a URL and a description of the columns
Browse AI Table Studio takes a URL, proposes tables and columns with sample rows and lets you approve each column before saving (https://help.browse.ai/en/articles/10486935-how-browse-ai-uses-ai). Robots has no AI in the builder; the backlog idea for natural-language act and extract steps is about runtime steps. First scope: a 'Start from a URL' dialog that loads the page in the runner, sends a trimmed DOM to the AI gateway and returns goto + extract_list (with fields and Next pagination) as unsaved steps the user can run and edit. Priority: high. Effort: M. Compared with Browse AI.
- 0
Robots: Pick what to extract by clicking on a live preview of the page
Browse AI builds robots by pointing at the page: Capture List and Capture Text in its Chrome extension (https://help.browse.ai/en/articles/10492719-how-to-build-a-robot-using-the-chrome-extension). Robots asks for CSS selectors typed by hand, and the bookmarklet recorder only captures clicks, fills, selects and keys, not extractions (lib/robots-recorder.ts). A user who does not know CSS cannot build the first robot. First scope: a 'Pick on page' button in Extract and Extract list that opens a sandbox screenshot with element boxes; clicking a card proposes the card selector and its fields with generateSelector, then the same mode inside the bookmarklet. Priority: high. Effort: L. Compared with Browse AI.
- 0
Robots: Export run tables to CSV and prune old runs with a retention policy
Browse AI lets users download captured lists and keeps task data for a plan-based retention window (https://www.saaspricepulse.com/tools/browse-ai), and Browserbase states 7-day data retention on its free plan (https://docs.browserbase.com/guides/plans-and-pricing). Our RunsPanel can show output as a table but only offers copying the JSON (components/robots/RunsPanel.tsx), and nothing ever deletes robot_runs rows, which store up to 200 KB of output and 600 log entries each, so a 15-minute schedule grows the table without bound (lib/robots.ts, lib/schema/migrations.ts). First scope: a 'Download CSV' button per table variable, and a cron cleanup that keeps the last 200 runs or 90 days per robot (configurable via env) while keeping the latest success. Priority: low. Effort: S. Compared with UiPath (StudioX, Autopilot, Orchestrator, Healing Agent), Browse AI, Axiom.ai, Bardeen, Browserbase/Stagehand, Skyvern.
- 0
Robots: Let Agents use robots as tools to browse and act on websites
Browserbase and Stagehand are built so AI agents can drive a real browser (https://docs.stagehand.dev/), and Skyvern runs LLM-driven workflows with login blocks (https://skyvern.com/docs/developers/features/authentication-and-2fa). Our Agents product has no browser or robot tool: nothing under lib/agent* or app/api/agents references robots, and startRobotRun is called only from the run route and the Flow engine. Exposing existing robots (with their typed inputs) as agent tools turns Robots into the 'hands' of the suite without building a general browser agent. First scope: agent setting to attach up to 5 robots; each becomes a tool whose JSON schema is generated from robotInputSchema, runs with trigger 'agent', and returns output vars plus screenshot URLs. Priority: low. Effort: M. Compared with UiPath (StudioX, Autopilot, Orchestrator, Healing Agent), Browse AI, Axiom.ai, Bardeen, Browserbase/Stagehand, Skyvern.
- 0
Robots: Expose a public API to run robots and bulk-run one robot over many inputs
Browse AI has a REST API with tasks history and POST /v2/robots/{robotId}/bulk-runs that queues up to 1,000 input sets per call (https://www.browse.ai/monitoring), and Axiom.ai has a developer API to trigger bots (https://axiom.ai/docs/developer-hub/api/reference/errorcodes/). Our run endpoint requires an admin session cookie (app/api/robots/[id]/run/route.ts, canEditRobots in lib/robots-http.ts) and app/api/v1 has agents, calendar, crm and links but no robots. Teams want to run a robot from their own scripts, Zapier or a spreadsheet of 200 URLs. First scope: /api/v1/robots/{id}/runs (POST with inputs, GET list) and /api/v1/runs/{id} using existing API keys, plus a bulk endpoint that enqueues runs once the queue from the busy fix exists; document in openapi.json. Priority: medium. Effort: M. Compared with UiPath (StudioX, Autopilot, Orchestrator, Healing Agent), Browse AI, Axiom.ai, Bardeen, Browserbase/Stagehand, Skyvern.
- 0
Robots: Add natural-language act and extract steps backed by the AI gateway with caching
Stagehand exposes act(), extract() with a schema and observe(), caches resolved actions keyed by instruction and page content, and self-heals when a cached selector stops resolving (https://docs.stagehand.dev/v4/basics/act, https://docs.stagehand.dev/v4/best-practices/caching); UiPath Autopilot generates StudioX workflows from a plain-language description (https://docs.uipath.com/studiox/standalone/2024.10/release-notes/2024-10-1). Our steps all require CSS selectors or exact visible text and there is no AI step type (lib/robots-schema.ts STEP_TYPES). AI steps lower the skill bar for non-technical members and fix brittle selectors. First scope: an 'ai_extract' step (instruction plus field list, returns rows) and an 'ai_click' step that resolves an element from an accessibility snapshot once, caches the chosen selector on the robot, and re-resolves only on failure. Priority: medium. Effort: L. Compared with UiPath (StudioX, Autopilot, Orchestrator, Healing Agent), Browse AI, Axiom.ai, Bardeen, Browserbase/Stagehand, Skyvern.
- 0
Robots: Replace the robot id text box in the Flow node with a picker that loads inputs
UiPath Orchestrator and Axiom.ai let you pick the automation to trigger and show its parameters (https://docs.uipath.com/orchestrator/docs/alerts, https://www.axiom.ai/docs/no-code-tool/how-it-works/loop). Our app.robot_run node defines robotId as a plain text field with placeholder 'robot id' and inputs as free key/value pairs (lib/flow-schema.ts), and nothing in components/flow resolves robots, while sub-flows get a flowRef picker. Users must copy a UUID from the URL and guess input names, so 'trigger a robot from Flow' is technically present but hard to use. First scope: a robotRef field kind that lists workspace robots and prefills the inputs kv with the robot's declared input names and required flags, plus outputs autocompletion for {{steps.x.output.var}}. Priority: medium. Effort: S. Compared with UiPath (StudioX, Autopilot, Orchestrator, Healing Agent), Browse AI, Axiom.ai, Bardeen, Browserbase/Stagehand, Skyvern.
- 0
Robots: Add a TOTP secret to credentials so robots can log in behind 2FA
Skyvern stores an authenticator key on a password credential and generates a fresh TOTP code at login, using the next code if fewer than 20 seconds remain (https://www.skyvern.com/docs/credentials/totp/, https://skyvern.com/docs/developers/features/authentication-and-2fa). Our credential types are api_key, basic, oauth_token and app tokens with no TOTP field (lib/flow-credentials-schema.ts), and the recorder deliberately masks one-time-code fields (lib/robots-recorder.ts), so a 2FA login can only work while a persisted session survives. Most admin portals small teams automate (banks, tax sites, supplier portals) require 2FA. First scope: optional totpSecret (secret) on the basic type and a {{credentials.NAME.totp}} template value computed at use time inside the interpreter, redacted like other secrets. Priority: medium. Effort: S. Compared with UiPath (StudioX, Autopilot, Orchestrator, Healing Agent), Browse AI, Axiom.ai, Bardeen, Browserbase/Stagehand, Skyvern.
- 0
Robots: Upsert extracted rows into CRM contacts and companies with mapping and dedupe
Bardeen positions scraping straight into GTM tools (scrape, enrich, export to CRM) with credits per scraped row (https://www.bardeen.ai/pricing), and Axiom.ai writes scrape results directly to Google Sheets (https://axiom.ai/guides/scrape-links-from-a-website/). Today a robot table reaches our CRM only by building a Flow with app.crm_contact per row, and the only robot template in Flow is robotToSheets (lib/flow-schema.ts). Robot output landing as deduplicated CRM records is something no standalone scraper can do natively and exploits our suite. First scope: a robot 'Destinations' panel that maps one table variable to CRM contacts or companies (column to field, dedupe by email or domain, source tag = robot name), applied after each successful run, with counts created/updated in the run view. Priority: medium. Effort: M. Compared with UiPath (StudioX, Autopilot, Orchestrator, Healing Agent), Browse AI, Axiom.ai, Bardeen, Browserbase/Stagehand, Skyvern.
- 0
Robots: Detect changes between runs and emit a robot.output.changed event with a diff
Browse AI monitors re-run a robot on a schedule and alert by email or webhook when captured text, lists or a screenshot change beyond a 1/5/10% threshold (https://www.browse.ai/monitoring, https://intercom.help/browse-ai-089fe61c5dc3/en/articles/10535487-how-can-i-setup-monitoring-for-changes). Our price-check template runs daily but nothing compares the output to the previous run; the only events are finished and failed (lib/robots.ts, lib/flow-schema.ts). Competitor price, job board and stock monitoring is the most common small team use case and it needs 'tell me only when it changed'. First scope: per-robot 'watch variables' list, compare against the last successful run output, store a diff (added/removed rows, changed values) on the run, and emit robot.output.changed for Flow and webhooks. Priority: medium. Effort: M. Compared with UiPath (StudioX, Autopilot, Orchestrator, Healing Agent), Browse AI, Axiom.ai, Bardeen, Browserbase/Stagehand, Skyvern.
- 0
Robots: Add continue-on-error per step and per loop item with a warnings status
Axiom.ai offers Continue on error, a Try/Catch step, and loops that log a warning for a failing row instead of stopping, plus 'succeeded with warnings' notifications (https://axiom.ai/docs/no-code-tool/reference/settings/error-handling/continue-on-error/, https://axiom.ai/blog/getting-to-know-the-trycatch-step/). In our runList any failed step throws and aborts the whole run, including one bad item inside a loop_over of 50 URLs (lib/robots-interpreter.ts), and RobotRunStatus only has success and failed. Losing 49 good results because of one broken page makes loop robots unusable. First scope: onError: 'fail' | 'continue' on leaf steps and loop_over, collect per-item errors into vars.errors, and a new 'partial' run status shown in RunsPanel and emitted in robot.run.finished. Priority: medium. Effort: S. Compared with UiPath (StudioX, Autopilot, Orchestrator, Healing Agent), Browse AI, Axiom.ai, Bardeen, Browserbase/Stagehand, Skyvern.
- 0
Robots: Add pagination that clicks Next until missing and appends rows across pages
Axiom.ai has a Find pager setting and a scrape API that keeps paginating until max results or the pager selector stops matching (https://axiom.ai/scrape/multiple-pages/, https://axiom.ai/docs/developer-hub/api/step-functions/scrape/). Our only loop is loop_over over a known list, nested loops are rejected, and if_exists can only jump forward (lib/robots-schema.ts validateSteps), so a 'repeat until there is no Next button' pattern is impossible. Multi-page results are the default for directories, marketplaces and admin panels. First scope: a paginate option on extract_table and extract_list (next selector, max pages up to 20, wait after click) that concatenates rows into the same variable, counted against MAX_PAGES and MAX_EXECUTED_STEPS. Priority: high. Effort: M. Compared with UiPath (StudioX, Autopilot, Orchestrator, Healing Agent), Browse AI, Axiom.ai, Bardeen, Browserbase/Stagehand, Skyvern.
- 0
Robots: Add an extract_list step that turns repeating cards into rows with named fields
Browse AI 'capture list' and Bardeen scrapers extract repeating items (product cards, search results, directory entries) into rows with columns the user names (https://help.browse.ai/en/articles/10486935-how-browse-ai-uses-ai, https://www.bardeen.ai/pricing), and Stagehand extract() returns typed data validated against a schema (https://docs.stagehand.dev/). We have extract_table, which only works on HTML <table>, and extract mode all, which returns a flat array of texts for one selector (lib/robots-interpreter.ts). Most real scraping targets (listings, leads, prices) are div-based lists, so today users must write an evaluate step in JS. First scope: extract_list with an item selector plus up to 20 fields (name, relative selector, text or attribute), capped at 500 rows, rendered by the existing table view in RunsPanel. Priority: high. Effort: M. Compared with UiPath (StudioX, Autopilot, Orchestrator, Healing Agent), Browse AI, Axiom.ai, Bardeen, Browserbase/Stagehand, Skyvern.
- 0
Robots: Record fallback selectors per step and self-heal when the primary one breaks
UiPath Unified Target keeps several targeting strategies and the Healing Agent votes on a candidate selector when one is not found, recommending or applying fixes (https://docs.uipath.com/agents/automation-cloud/latest/user-guide-ha/what-is-healing-agent); Browse AI says it detects layout changes and adapts the robot, notifying the user (https://help.browse.ai/en/articles/10486935-how-browse-ai-uses-ai); Stagehand re-infers a selector when a cached one stops resolving (https://docs.stagehand.dev/v4/best-practices/caching). Our recorder computes candidates but keeps only the first unique one (generateSelector in lib/robots-recorder.ts), and resolveTarget tries one selector and then visible text (lib/robots-interpreter.ts). Selector rot is the main maintenance cost of browser robots. First scope: store a fallbacks array (testid, aria-label, name, text, CSS path) per step from the recorder, try them in order at runtime, log 'healed via fallback N', and offer a one-click 'promote fallback' in the run view. Priority: high. Effort: M. Compared with UiPath (StudioX, Autopilot, Orchestrator, Healing Agent), Browse AI, Axiom.ai, Bardeen, Browserbase/Stagehand, Skyvern.
- 0
Robots: Capture a screenshot, URL and DOM snippet automatically when a step fails
Browserbase records every session as video with a Session Inspector and live debug URL so failures can be replayed (https://docs.browserbase.com/features/session-recording, https://docs.browserbase.com/features/session-debugger). Our interpreter only takes screenshots when the robot has an explicit screenshot step; on a failed step it logs a code like s3: action_timeout plus the first line of the Playwright message (lib/robots-interpreter.ts runList). Without seeing the page at the moment of failure (cookie banner, login wall, changed layout) users cannot fix a robot. First scope: in the catch of runList, take one failure screenshot (outside the 5-shot cap), record page.url() and the outerHTML of the nearest match or body excerpt, and show them next to the failed log row in RunsPanel. Priority: high. Effort: S. Compared with UiPath (StudioX, Autopilot, Orchestrator, Healing Agent), Browse AI, Axiom.ai, Bardeen, Browserbase/Stagehand, Skyvern.
- 0
Robots: Send failure alerts and auto-pause schedules after N consecutive failures
Axiom.ai notifies by email or webhook when a run fails, succeeds, or succeeds with warnings (https://axiom.ai/docs/troubleshooting/error-handling), and UiPath Orchestrator sends Error alerts for failed jobs and disables a trigger after its job fails five times in a row (https://docs.uipath.com/orchestrator/standalone/2023.4/user-guide/alerts). We only emit robot.run.failed (lib/robots.ts); there is no in-app or email notification unless the user builds a Flow, and lib/notifications.ts has no robot hook. A scheduled robot that silently fails for a week is the top reason people stop trusting RPA. First scope: per-robot setting 'notify on failure' (creator and admins, in-app plus email, digest at most once per hour) and an auto-pause after 5 consecutive scheduled failures with a banner in RobotsHome. Priority: high. Effort: M. Compared with UiPath (StudioX, Autopilot, Orchestrator, Healing Agent), Browse AI, Axiom.ai, Bardeen, Browserbase/Stagehand, Skyvern.
- 0
Robots: Queue runs instead of failing as busy when another workspace run is active
UiPath Orchestrator queues pending jobs per robot and only raises an alert when a trigger cannot create jobs (https://docs.uipath.com/orchestrator/standalone/2023.4/user-guide/alerts), and Browse AI queues tasks for bulk runs (https://www.browse.ai/monitoring). Our startRobotRun inserts a run only when no other run in the workspace is running, otherwise returns busy (lib/robots.ts). For schedules, runDueRobots advances next_run_at before starting, so a collision with a manual or Flow run records a failed busy run and that scheduled execution is lost; from Flow, app.robot_run fails unless the node has retries (default 0, lib/flow-engine.ts robotNode). Teams with two or three robots will see random failures they did not cause, which destroys trust. First scope: add a queued status, insert as queued when busy, and drain the queue in /api/cron/robots and after each run finishes, with a max queue age. Priority: high. Effort: M. Compared with UiPath (StudioX, Autopilot, Orchestrator, Healing Agent), Browse AI, Axiom.ai, Bardeen, Browserbase/Stagehand, Skyvern.
- 0
Robots: browser automations
Record steps in the browser and run them on a schedule or from Flow.