Skip to content
Omnibot Omnibot
Docs
Blog

Roadmap

Omnibot Product Roadmap

Last updated: 2026-07-07

This document is the current product-status source of truth. Historical implementation plans in docs/superpowers/plans/ remain useful for design intent and task history, but their unchecked boxes may not reflect the current codebase.

For a reader-friendly overview of the current product, see Product Overview.

Product Direction

Omnibot focuses on letting AI agents read, inspect, debug, and operate the user’s real Chromium browser: real tabs, real login state, real extension environment, and explicit tab targeting through a local daemon plus CLI.

Near-term product priorities:

  • Keep the local real-browser workflow stable and explicit.
  • Make common browser automation tasks possible without raw JavaScript.
  • Provide clear fallback tiers when semantic or ref-based operations fail.
  • Preserve agent-friendly stdout while keeping diagnostics on stderr.
  • Package skills and docs so agents can reliably discover the correct workflow.

Status Legend

  • Done: implemented in CLI/action/daemon surface and covered by unit or E2E contracts.
  • Partial: command surface exists, but production behavior has known limits or only a foundational path is implemented.
  • Experimental: available for targeted testing, not yet a default production workflow.
  • Planned: not implemented or only represented by design notes.

Done

Runtime And Distribution

  • Local daemon and CLI v2 runtime: omnibot doctor, status, start, stop, run, daemon auto-start, and local ports 18764/18765.
  • Browser extension transport: WebSocket primary path, HTTP long-poll fallback, tab/session tracking, extension reconnect behavior, and tab update broadcasting.
  • npm packaging: platform packages, CLI package, packaged skills path/install commands, and version checks.
  • License lifecycle: license status, local persisted activation state, usage count, and extension status push/refresh.
  • Explicit tab targeting: page-state commands require --tab-id; no implicit default target for browser state changes.

Reading And Page Understanding

  • read: clean rendered page reading with --screens, URL temporary-tab mode, tab-targeted mode, and JSON/text output.
  • snapshot: accessibility tree snapshot, @eN refs, interactive filtering, compact/depth/url options, selector scoping, rich text editor controls, popup/modal controls, and combobox option probing.
  • get: title, URL, text, HTML, value, attribute, count, box, and style reads.
  • is: visible, enabled, and checked state checks.
  • find: semantic locators for role, text, label, placeholder, alt, title, testid, and nth, with subactions such as click, fill, type, hover, focus, check, uncheck, and text.

User-Like Browser Operations

  • Ref/selector operations: click, dblclick, fill, type, press, hover, focus, select, check, uncheck, scroll, scrollintoview, and element-level drag.
  • Upload: CDP DOM.setFileInputFiles, DOM nodeId fallback, and JS DataTransfer fallback are implemented and unit-covered.
  • Navigation and tab lifecycle: navigate, open, goto, close, tab list/new/close, window new, frame, back, forward, reload, and pushstate.
  • Wait and batch: selector/text/URL/load/function/fixed-time waits, hidden/visible state, and JSON/file batch commands.

Debugging And Evidence

  • Screenshots: normal/full/annotated screenshots, PNG/JPEG options, base64/file output paths.
  • Console: logs, errors command path, and clear.
  • Dialogs: JavaScript dialog logs, clear, accept/dismiss, and prompt text handling.
  • Network: extension-backed capture start/stop/clear/logs/summary.
  • Raw CDP: cdp <method> <params-json> for last-resort inspection.

Fallback And Advanced Operation Surfaces

  • CUA coordinate mouse: mouse click, mouse move, mouse scroll, and mouse drag command surface backed by CDP input events.
  • DomCUA: dom visible, dom click, dom dblclick, and dom scroll for visible DOM node-id fallback workflows.
  • Clipboard: clipboard read and clipboard write through page context.
  • Viewport: viewport get and viewport set via page state/CDP emulation.
  • Assets: assets list and assets export; export writes asset metadata into a zip when an output path is provided.
  • Browser/session metadata: browser list/current/claim/release, session name/list.
  • Site-skill exploration: explore run/list/show/promote foundation for generated site skills and operation memory.

Partial Or Experimental

Human Verification

  • verify inspect detects and extracts metadata for NetEase Yidun captcha widgets.
  • It does not solve captchas, does not provide a production-grade bypass workflow, and requires visual cross-checking for state and coordinate mapping.

Mouse Drag Realism

  • mouse drag --fast is the stable linear path.
  • The default realistic drag trajectory is available but experimental; timing, jitter, and success rates still need production validation.

Console And Network Capture Completeness

  • Console and network command surfaces exist and have extension-backed capture paths.
  • console errors currently shares the console log collection path; callers should verify level filtering in output.
  • Network and dialog capture require a current browser extension; older extensions return explicit upgrade/reload errors.

Record, Replay, And Trace

  • record start/stop, replay, and trace start/stop are exposed and have basic payload/file shapes.
  • Automatic capture of every CLI action into recorded_actions / trace_events is not yet wired as a complete product workflow.
  • Replay supports a limited action set and should be treated as a foundation, not stable browser-flow automation.

Visibility Modes

  • visibility status and visibility set model visible, background, dedicated-profile, and headless modes.
  • visibility launch currently returns status: planned for dedicated/headless launch; it does not yet spawn or attach a separate browser profile.

Assets Export

  • assets list/export collect resource metadata from the page.
  • Full offline page mirroring, binary asset download, and HAR-quality archive export are not implemented.

Planned Or Not Implemented

Automation Parity Gaps

  • Full Playwright-compatible locator/action parity beyond the current semantic locator subset.
  • Complete DomCUA node lifecycle with stable IDs across DOM mutations and frames.
  • Full CUA policy layer, confirmation workflow, and action safety policy.
  • Robust cross-origin iframe introspection beyond current best-effort behavior.

Browser Capability Gaps

  • Dedicated automation browser launch and attach flow for visibility launch.
  • Headless Chromium mode with extension compatibility guarantees.
  • Download management, PDF export, full page archive, and full HAR export.
  • Persistent cookie/storage/state vault and auth profile management.
  • Full clipboard/filesystem workflows beyond current text clipboard and file upload support.

Debugging And Observability Gaps

  • Trace viewer/dashboard.
  • Record/replay as a durable no-code macro workflow.
  • Profiler, Web Vitals, React/Vue component introspection, and performance timelines.
  • Streaming event feed for long-running operations.

Product And Platform Gaps

  • Remote multi-user daemon as a supported product mode.
  • Cloud browser providers, Lightpanda, iOS/Appium, and mobile automation.
  • Team dashboard, policy administration, billing/payment integration, and hosted account management.
  • AI-assisted captcha solving. Current scope is inspection only.

Suggested Next Milestones

M1: Stabilize Current Local Browser Surface

  • Keep snapshot + click/fill/type/wait as the primary happy path.
  • Expand E2E coverage for real browser workflows, especially popup controls, rich text editors, upload, network/dialog capture, and explicit tab targeting.
  • Make command reference and skills match the actual CLI command surface on every release.

M2: Finish Recording And Trace Workflow

  • Wire action dispatch to append record and trace events consistently.
  • Define a stable replay schema and support more actions.
  • Add trace export validation and a minimal human-readable trace report.

M3: Dedicated Profile And Visibility Launch

  • Implement actual browser launch for dedicated-profile mode.
  • Document login-state limitations and extension loading requirements.
  • Decide whether headless mode remains experimental or becomes a supported non-user-browser mode.

M4: Observability And Archive Depth

  • Improve console/network filtering and retention.
  • Add richer asset export and HAR-style summaries.
  • Evaluate whether trace viewer/dashboard belongs in CLI-only form or a separate UI.

M5: Platform Expansion

  • Revisit remote daemon, hosted/team workflows, policy controls, and cloud browser providers only after local real-browser automation remains stable across release gates.