47 of 51 criteria have a sourced finding; 41 were also cross-checked by a second agent. Every comparison that includes Omnigent links here for scope, limitations, atomic answers and sources.
Humans approve the rubric, templates and prototype; they do not review every finding. An agent source cross-check is not a hands-on product test. A missing sourced finding is not evidence that a capability is absent.
Editor extensions [partial] A repository VS Code extension (v0.1.0) only opens the running local server's UI in an editor pane; sessions, diffs and remote servers are out of scope and Marketplace publication was not verified. The web and desktop Files panel opens files in VS Code or Cursor. Sources: omnigent:nov21-vscode-src, omnigent:nov21-webui-sessions.
Named coding harnesses support direct/native modes with distinct support levels. Omnigent v0.21.0 ships the antigravity-native harness, launched with omnigent antigravity or omnigent agy, with CLI/web setup and approval handling.
Scope
Surfaces: desktop, web, self-hosted · Version: Retained findings from first-party docs and repository read 2026-11-07; v0.21.0 source (commit 2f636330) read 2026-11-10; v0.22.0 source (commit a9a0a163) read 2026-11-16; omnigent.ai docs read 2026-11-21. No installed release tested. · Plan: Not specified · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-07 (15 days old)
Agent cross-check
2026-11-22
Every atomic question has a cited first-party answer.
Antigravity is community-supported and requires the native runtime and configured credentials; direct and terminal modes differ.
Model providers [supported] API keys (Anthropic, OpenAI, OpenRouter, Groq, DeepSeek, xAI, Mistral, Together, Fireworks), Claude or ChatGPT subscriptions through their CLIs, any OpenAI- or Anthropic-compatible gateway or local server (LiteLLM, Azure OpenAI, Ollama, vLLM and others), or Databricks. Sources: omnigent:nov21-models, omnigent:nov21-faq.
Use native Claude/ChatGPT subscriptions, API credentials or compatible gateways.
Scope
Surfaces: desktop, web, self-hosted · Version: Retained findings from first-party docs and repository read 2026-11-07; omnigent.ai docs read 2026-11-21. No installed release tested. · Plan: Not specified · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-07 (15 days old)
Agent cross-check
2026-11-22
Every atomic question has a cited first-party answer.
7 atomic answers
Agent subscriptions [supported] Claude Pro/Max and ChatGPT plan via official claude/codex CLI login. Sources: omnigent:nov07-readme.
Api accounts [supported] Own Anthropic/OpenAI/provider keys. Sources: omnigent:nov07-readme.
Vendor credits [unsupported] Omnigent sells no credits: users bring their own keys, subscriptions, gateways or Databricks workspace. Sources: omnigent:nov21-faq, omnigent:nov21-models.
Local models [supported] Ollama/vLLM compatible gateway route. Sources: omnigent:nov07-readme.
Account switching [supported] Setup manages credentials grouped by agent; defaults per agent. Sources: omnigent:nov07-readme.
No git [unknown] No git: no explicit product-scoped answer found in the checked first-party material.
Local git without remote [unknown] Local git without remote: no explicit product-scoped answer found in the checked first-party material.
Hosted git [supported] GitHub PRs get a tracked Pull Requests tab with edit, review and merge actions through gh; GitLab merge requests and Azure DevOps pull requests are read-only. Sources: omnigent:nov21-webui-github.
Sdk [supported] Python omnigent_client SDK from the omnigent-client package, version matched to the server. Sources: omnigent:nov21-programmatic.
Mcp server [unknown] Mcp server: no explicit product-scoped answer found in the checked first-party material.
Webhooks [unsupported] Trigger templates poll services and do not receive webhooks. A v0.23.0 source search found only internal harness callback routes in routes_hooks.py, hidden from the public API, and no user-configurable outgoing or incoming webhook. Sources: omnigent:nov21-webui-automations, omnigent:ade157-src-webhooks.
Plugins [supported] Python packages can add harnesses through the omnigent.community.harness entry point (direct and headless only) and add cloud sandbox providers. Sources: omnigent:nov21-harness-community, omnigent:nov21-sandbox-host.
Mobile implementation [supported] The iOS (Swift) and Android (Kotlin) apps are thin WebView shells around the server-served web UI at v0.23.0. Sources: omnigent:nov07-mobile, omnigent:nov21-src-clients.
Implementation evidence [supported] Pinned v0.23.0 source: React and Vite web UI, Electron 45.2.1 desktop, and Swift (iOS) and Kotlin (Android) WebView shells around the server-served UI. Sources: omnigent:nov21-src-clients.
External import [partial] Polling trigger templates start one session per new Jira ticket, Slack message, GitHub item, Linear issue or PagerDuty incident (up to 10 per run); the agent needs that service's tools. Sources: omnigent:nov21-webui-automations.
Acceptance criteria [unknown] Acceptance criteria: no explicit product-scoped answer found in the checked first-party material.
Editors [unknown] Editors: no explicit product-scoped answer found in the checked first-party material.
Revisions [unknown] Revisions: no explicit product-scoped answer found in the checked first-party material.
Implementation links [partial] Polly Agents panel links the orchestrator to delegated sub-agent sessions and each coding session tracks its own PRs; a persistent plan-to-PR link is not documented. Sources: omnigent:nov07-polly, omnigent:nov07-web.
Amendment [partial] Inline comments on a plan file are sent back with Address All for revision; there is no separate plan object with revisions. Sources: omnigent:nov21-webui-comments.
Multimodal inputs [supported] Web composer accepts pasted images and attached PDFs, text/code, archives, Office documents and SQLite files; some types need Claude Code or Codex. Sources: omnigent:nov07-desktop, omnigent:nov07-readme, omnigent:nov07-web.
Freshness [unknown] Freshness: no explicit product-scoped answer found in the checked first-party material.
The scoped native UI and database have agent sessions, project grouping and scheduled automation definitions. Bounded inspection found no independent native issue/backlog/to-do work-item list; scheduled trigger definitions and run statuses are separate.
Scope
Surfaces: desktop, web, self-hosted · Version: V0.21.0 source (commit 2f636330) read 2026-11-13; v0.23.0 source (commit 033963227a, released 2026-11-19) read 2026-11-21; omnigent.ai docs read 2026-11-21. No installed release tested. · Plan: Not specified · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-13 (9 days old)
Agent cross-check
2026-11-22
Only the named atomic answers are established by this source pass. Remaining atomic answers retain their documented gaps.
17 atomic answers
Create [unsupported] Native stores/routes expose sessions, projects and schedules rather than issue/backlog work-item creation. Sources: omnigent:gap39-native-work.
Edit [unknown] Edit: no explicit product-scoped answer found in the checked first-party material.
Discuss [unknown] Discuss: no explicit product-scoped answer found in the checked first-party material.
Prioritize [unknown] Prioritize: no explicit product-scoped answer found in the checked first-party material.
Label [unknown] Label: no explicit product-scoped answer found in the checked first-party material.
Assign [unknown] Assign: no explicit product-scoped answer found in the checked first-party material.
Reopen [unknown] Reopen: no explicit product-scoped answer found in the checked first-party material.
Project epic views [partial] Projects group sessions in non-nesting folders with project defaults; no epic view. Sources: omnigent:nov21-webui-sessions.
Idea parking [unknown] Idea parking: no explicit product-scoped answer found in the checked first-party material.
Explicit dispatch [unknown] Explicit dispatch: no explicit product-scoped answer found in the checked first-party material.
Task session separation [unsupported] The native work record is a session; the separate task store contains scheduled trigger definitions, not backlog work items. Sources: omnigent:gap39-native-work.
Metadata [unknown] Metadata: no explicit product-scoped answer found in the checked first-party material.
Durable states [unknown] Durable states: no explicit product-scoped answer found in the checked first-party material.
External tracker requirement [partial] No native issue tracker; issue intake relies on Jira, GitHub, Linear or other service tools through trigger templates. Sources: omnigent:nov21-webui-automations.
Agents create child sessions and read other sessions' status and history through built-in tools; Omnigent has no issue objects, so issue claims, assignment, states and dependencies do not apply to sessions.
Scope
Surfaces: desktop, web, self-hosted · Version: Omnigent.ai docs read 2026-11-21. No installed release tested. · Plan: Not specified · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
unknown
Delivery
unknown
Checked
2026-11-21 (1 day old)
Agent cross-check
2026-11-22
Sessions, not issues, are the durable work items; no issue IDs, fields, claims, states or dependencies exist for agents to operate on.
Only documented tool behavior was checked; no multi-agent fixture was run.
14 atomic answers
Shared ids [unknown] Shared ids: no explicit product-scoped answer found in the checked first-party material.
Read [partial] Agents read other sessions' status and recent history with sys_session_get_info and sys_session_get_history. Sources: omnigent:nov21-programmatic.
Create [partial] Agents create child sessions with sys_session_create; sessions are the work items and there is no issue object. Sources: omnigent:nov21-programmatic.
Update [unknown] Update: no explicit product-scoped answer found in the checked first-party material.
Claim [unknown] Claim: no explicit product-scoped answer found in the checked first-party material.
Assign [unknown] Assign: no explicit product-scoped answer found in the checked first-party material.
State changes [unknown] State changes: no explicit product-scoped answer found in the checked first-party material.
Dependencies [unknown] Dependencies: no explicit product-scoped answer found in the checked first-party material.
Children [unknown] Children: no explicit product-scoped answer found in the checked first-party material.
Results [unknown] Results: no explicit product-scoped answer found in the checked first-party material.
Human visibility [unknown] Human visibility: no explicit product-scoped answer found in the checked first-party material.
Attribution [unknown] Attribution: no explicit product-scoped answer found in the checked first-party material.
Permissions [unknown] Permissions: no explicit product-scoped answer found in the checked first-party material.
Conflicts [unknown] Conflicts: no explicit product-scoped answer found in the checked first-party material.
Quotas [partial] spawn_bounds limits sub-agent dispatches per turn (default 5) and user budgets cap spend; no global session or concurrency quota is documented. Sources: omnigent:nov21-policies.
Observed concurrency [unknown] Observed concurrency: no explicit product-scoped answer found in the checked first-party material.
Polly routes implementation and review work to child agents.
Scope
Surfaces: desktop, web, self-hosted · Version: Retained findings from first-party docs and repository read 2026-11-07; omnigent.ai docs read 2026-11-21. No installed release tested. · Plan: Not specified · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-07 (15 days old)
Agent cross-check
2026-11-22
Not established by this source pass: claims, queues, duplicate_prevention.
Answered only in part, with the limit stated in each atomic answer: human_assignment.
7 atomic answers
Human assignment [partial] An owner shares a session with Read or Edit access so a teammate can co-drive; there is no assignee field or queue. Sources: omnigent:nov21-collaborate.
Claims [unknown] Claims: no explicit product-scoped answer found in the checked first-party material.
Queues [unknown] Queues: no explicit product-scoped answer found in the checked first-party material.
Routing [supported] Cross-vendor reviewer differs from implementing agent. Sources: omnigent:nov07-polly.
Duplicate prevention [unknown] Duplicate prevention: no explicit product-scoped answer found in the checked first-party material.
Ownership enforcement [supported] The server enforces owner-only control of permission mode, launch args, raw shell, absolute host paths and the session's agent; Edit collaborators are refused. Sources: omnigent:nov21-collaborate.
Followup [supported] sys_session_send sends follow-ups to the caller's direct children; reusing a named sub-agent and title continues that child. Sources: omnigent:nov21-programmatic.
Persistent recurring automations launch agent sessions from server.
Scope
Surfaces: desktop, web, self-hosted · Version: Retained findings from first-party docs and repository read 2026-11-07; omnigent.ai docs read 2026-11-21. No installed release tested. · Plan: Not specified · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-07 (15 days old)
Agent cross-check
2026-11-22
Not established by this source pass: completion_conditions.
Answered only in part, with the limit stated in each atomic answer: events, client_independence.
8 atomic answers
Schedules [supported] Minimum ten-minute interval on connected hosts or continued sessions; one hour for new sandbox sessions. Sources: omnigent:nov07-automation.
Events [partial] Trigger templates for Jira, Slack, GitHub, Linear and PagerDuty poll for new items (10 minutes by default with automations v2, else hourly) and need the service's tools; no webhooks. Sources: omnigent:nov21-webui-automations.
Completion conditions [unknown] Completion conditions: no explicit product-scoped answer found in the checked first-party material.
Scheduler location [supported] Server scheduler; selected execution host must be available. Sources: omnigent:nov07-automation.
Overlaps [supported] Overlapping fires of the same task are skipped while the previous firing is still dispatching. Sources: omnigent:nov07-automation.
Client independence [partial] The server schedules runs. A connected-host run needs that host online or it is recorded as failed; a managed-sandbox run provisions a fresh sandbox and needs no client. Sources: omnigent:nov21-scheduled-tasks, omnigent:nov21-sandbox-host.
Shared prompt/approval control and notifications are available across clients.
Scope
Surfaces: desktop, web, self-hosted · Version: Retained findings from first-party docs and repository read 2026-11-07; v0.21.0 release notes (2026-11-05); omnigent.ai docs read 2026-11-21. No installed release tested. · Plan: Not specified · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-07 (15 days old)
Agent cross-check
2026-11-22
Not established by this source pass: pause, stop, visible_errors.
Answered only in part, with the limit stated in each atomic answer: resume.
10 atomic answers
Pause [unknown] Pause: no explicit product-scoped answer found in the checked first-party material.
Stop [unknown] Stop: no explicit product-scoped answer found in the checked first-party material.
Resume [partial] Since v0.21.0 a session interrupted by a host restart recovers: the reply is marked interrupted and the next message runs without manual resume. The interrupted turn itself is not continued. Sources: omnigent:nov21-gh-v0210, omnigent:nov21-webui-sessions.
Visible errors [unknown] Visible errors: no explicit product-scoped answer found in the checked first-party material.
Chat terminal parity [supported] Terminal, web UI and desktop app are views of the same session; typing in one updates the others. Sources: omnigent:nov21-quickstart.
In the self-hosted v0.23.0 release, a separate server and hosts run sessions locally, on user-controlled remote machines or on cloud sandbox providers configured with the operator's own credentials; the open-source release provisions no vendor runtime. Databricks, which publishes Omnigent, also offers Omnigent on Databricks in Beta: a Databricks-operated server with Databricks Sandboxes, gated by workspace previews and limited to select AWS regions.
Scope
Surfaces: desktop, web, self-hosted · Version: Retained findings from first-party docs and repository read 2026-11-07; v0.21.0 source (commit 2f636330) read 2026-11-13; v0.23.0 source (commit 033963227a, released 2026-11-19) read 2026-11-21; omnigent.ai docs read 2026-11-21; Omnigent on Databricks (Beta) docs (updated 2026-09-28) read 2026-11-21. No installed release tested. · Plan: Self-hosted open-source release; Omnigent on Databricks (Beta) is preview-gated, limited to select AWS regions and not generally available. · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-07 (15 days old)
Agent cross-check
2026-11-22
Only the named atomic answers are established by this source pass. Remaining atomic answers retain their documented gaps.
Omnigent on Databricks (Beta) is a preview-gated, region-limited Beta without custom Python policies or bring-your-own model keys on Sandbox hosts; its billing and parity with the self-hosted release are not established.
Managed cloud [partial] The self-hosted v0.23.0 release provisions no vendor runtime: managed sandbox launchers run under the operator's server configuration and provider credentials. Separately, Omnigent on Databricks (Beta) is a managed edition from Databricks, the project's copyright holder, with a Databricks-operated server and Databricks Sandboxes. It needs workspace previews and is limited to select AWS regions. Sources: omnigent:gap39-microvm-code, omnigent:ade157-databricks-omnigent, omnigent:ade157-src-notice.
Git worktrees, remote sandbox containers and optional OS policies are separate isolation mechanisms. The optional microsandbox provider additionally gives each host a hardware-isolated microVM on supported macOS/Linux hosts.
Scope
Surfaces: desktop, web, self-hosted · Version: Retained findings from first-party docs and repository read 2026-11-07, 2026-11-13; v0.21.0 source (commit 2f636330) read 2026-11-13; omnigent.ai docs read 2026-11-21. No installed release tested. · Plan: Not specified · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-07 (15 days old)
Agent cross-check
2026-11-22
Only the named atomic answers are established by this source pass. Remaining atomic answers retain their documented gaps.
12 atomic answers
Checkout [unknown] Checkout: no explicit product-scoped answer found in the checked first-party material.
Branch [supported] The composer can name a worktree branch and base, or always start in a random worktree branch. Sources: omnigent:nov21-webui-sessions.
Vm [supported] The optional microsandbox provider boots each host in a libkrun microVM, with server-managed and CLI bootstrap modes. Requires Apple Silicon macOS or Linux KVM; the runtime is beta. Sources: omnigent:gap39-microvm, omnigent:gap39-microvm-code.
Secrets [supported] Omnibox gives agents placeholder tokens and a proxy injects real credentials only for allowed requests. Sources: omnigent:nov21-omnibox.
Code separation [unknown] Code separation: no explicit product-scoped answer found in the checked first-party material.
Server persists sessions through restarts; forks copy history with independent compute binding.
Scope
Surfaces: desktop, web, self-hosted · Version: Retained findings from first-party docs and repository read 2026-11-07; omnigent.ai docs read 2026-11-21. No installed release tested. · Plan: Not specified · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-07 (15 days old)
Agent cross-check
2026-11-22
Not established by this source pass: worker_crash, expiry, process_reconnect, state_reconstruction, duplicate_actions, manual_repair, multi_day_continuity.
Answered only in part, with the limit stated in each atomic answer: context_loss.
Ports [partial] Local sessions can open localhost apps in the desktop browser pane; port allocation and collisions are not documented. Sources: omnigent:nov21-gh-v0210.
Duplicate installs [unknown] Duplicate installs: no explicit product-scoped answer found in the checked first-party material.
Collisions [unknown] Collisions: no explicit product-scoped answer found in the checked first-party material.
Remote setup [supported] macOS/Linux desktop installs/authenticates/starts background host via SSH; not available on Windows. Sources: omnigent:nov07-deploy, omnigent:nov07-desktop.
Automatic repair [unknown] Automatic repair: no explicit product-scoped answer found in the checked first-party material.
Token limits [unknown] Token limits: no explicit product-scoped answer found in the checked first-party material.
Spend limits [supported] cost_budget ask thresholds; at max_cost_usd it blocks only turns/tool calls on expensive models (downgrade gate, not a hard stop). Sources: omnigent:nov07-policies.
Timeouts [unknown] Timeouts: no explicit product-scoped answer found in the checked first-party material.
Child cost [partial] subagent_cost_budget meters and gates a child subtree's spend; a per-child cost view is not documented. Sources: omnigent:nov21-policies.
Epic cost [unknown] Epic cost: no explicit product-scoped answer found in the checked first-party material.
Harness aggregation [supported] The usage page aggregates cost by harness and by model, and multi-harness sessions list every harness used. Sources: omnigent:nov21-cost-blog.
Billed vs estimated [partial] Costs are recorded from Omnigent's pricing catalog or custom pricing, not provider bills; unpriced tokens can run untracked. Sources: omnigent:nov21-spend-answer, omnigent:nov21-models.
Tokens [partial] Policy event context exposes input/output/total tokens and total_cost_usd to policies; a user-facing token view is not documented. Sources: omnigent:nov07-policies, omnigent:nov07-readme.
Subscription allowance [unknown] Subscription allowance: no explicit product-scoped answer found in the checked first-party material.
Provider reconciliation [unknown] Provider reconciliation: no explicit product-scoped answer found in the checked first-party material.
Compute cost [unsupported] Budgets and usage cover recorded LLM cost only and are documented as not an infrastructure budget. Sources: omnigent:nov21-spend-answer.
Refresh delay [unknown] Refresh delay: no explicit product-scoped answer found in the checked first-party material.
Revoke [unknown] Revoke: no explicit product-scoped answer found in the checked first-party material.
Tunnels [unknown] Tunnels: no explicit product-scoped answer found in the checked first-party material.
Defaults [supported] Approval guards for adding policies and creating or changing scheduled tasks are always on and cannot be removed. Sources: omnigent:nov21-policies.
Committed session events and authorship persist and export.
Scope
Surfaces: desktop, web, self-hosted · Version: Retained findings from first-party docs and repository read 2026-11-07; v0.23.0 source (commit 033963227a, released 2026-11-19) read 2026-11-21; omnigent.ai docs read 2026-11-21. No installed release tested. · Plan: Not specified · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-07 (15 days old)
Agent cross-check
2026-11-22
Not established by this source pass: approvals, edits, decisions, integrity.
Retention conflict: the 2026-09-03 blog says expired sessions are cleaned up automatically, but v0.23.0 code only marks them expired in a per-browser setting and deletes them on a manual Delete expired action.
Answered only in part, with the limit stated in each atomic answer: retention.
Retention [partial] At v0.23.0, archived sessions can be marked expired after 7, 30, 60 or 90 days in a per-browser setting and deleted manually with Delete expired; no audit retention is defined. Conflict: the 2026-09-03 blog describes automatic cleanup, which was not found in v0.23.0 code. Sources: omnigent:nov21-cost-blog, omnigent:nov21-src-archived-retention.
Integrity [unknown] Integrity: no explicit product-scoped answer found in the checked first-party material.
Changes and PR UI support diff review and inline GitHub conversations.
Scope
Surfaces: desktop, web, self-hosted · Version: Retained findings from first-party docs and repository read 2026-11-07; omnigent.ai docs read 2026-11-21. No installed release tested. · Plan: Not specified · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-07 (15 days old)
Agent cross-check
2026-11-22
Not established by this source pass: shipping_guidance, effort, missed_evidence, context_switching, roles, retained_context.
Answered only in part, with the limit stated in each atomic answer: pending_reviews.
13 atomic answers
Pending reviews [partial] The Inbox lists pending approvals and unread sessions and the PR tab lists tracked PRs; there is no queue of pending code reviews. Sources: omnigent:nov21-webui-comments, omnigent:nov21-webui-github.
Runtime context [partial] Co-drive keeps running on the owner's host; a shared link carries no uncommitted files elsewhere, but a fork onto a sandbox clones the source's recorded repositories. Sources: omnigent:nov21-handoff, omnigent:nov21-collaborate.
Owner disconnect [unknown] Owner disconnect: no explicit product-scoped answer found in the checked first-party material.
Ownership [unknown] Ownership: no explicit product-scoped answer found in the checked first-party material.
Attribution [unknown] Attribution: no explicit product-scoped answer found in the checked first-party material.
Context loss [unknown] Context loss: no explicit product-scoped answer found in the checked first-party material.
Harness transfer [supported] Clone session can choose different agent; bound agent cannot change in place. Sources: omnigent:nov07-collab, omnigent:nov07-web.
Polly routes worker changes to a different-vendor reviewer; policy gates are separately configured.
Scope
Surfaces: desktop, web, self-hosted · Version: Retained findings from first-party docs and repository read 2026-11-07; omnigent.ai docs read 2026-11-21. No installed release tested. · Plan: Not specified · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-07 (15 days old)
Agent cross-check
2026-11-22
Not established by this source pass: tests, lint, types, spec_checks, required_checks, failure_blocking, requirement_conflicts.
Answered only in part, with the limit stated in each atomic answer: ci, enforced_policy.
11 atomic answers
Tests [unknown] Tests: no explicit product-scoped answer found in the checked first-party material.
Lint [unknown] Lint: no explicit product-scoped answer found in the checked first-party material.
Types [unknown] Types: no explicit product-scoped answer found in the checked first-party material.
Ci [partial] The PR tab shows CI check results and a GitHub Actions workflow can run Omnigent PR reviews; checks are not a product delivery gate. Sources: omnigent:nov21-webui-github, omnigent:nov21-gh-actions.
Ci response [partial] CI checks are visible and refresh about every 30 seconds; no automatic fix-on-failure loop is documented. Sources: omnigent:nov21-webui-github.
Rebase [unknown] Rebase: no explicit product-scoped answer found in the checked first-party material.
Conflicts [unknown] Conflicts: no explicit product-scoped answer found in the checked first-party material.
Product state [supported] Checks/request state/merge actions; stale/incomplete data flagged. Sources: omnigent:nov07-pr, omnigent:nov07-web.
Product PR panel has merge controls; deployment authority remains unknown.
Scope
Surfaces: desktop, web, self-hosted · Version: Retained findings from first-party docs and repository read 2026-11-07; omnigent.ai docs read 2026-11-21. No installed release tested. · Plan: Not specified · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-07 (15 days old)
Agent cross-check
2026-11-22
Not established by this source pass: release, deploy.
Answered only in part, with the limit stated in each atomic answer: approve, exact_revision, authority_enforcement.
6 atomic answers
Approve [partial] Users request or withdraw reviewers, reply to reviews and see review states; submitting an approving review is not documented. Sources: omnigent:nov21-webui-github.
Release [unknown] Release: no explicit product-scoped answer found in the checked first-party material.
Deploy [unknown] Deploy: no explicit product-scoped answer found in the checked first-party material.
Exact revision [partial] Diff requests validate head/base; no digest-bound user approval contract established. Sources: omnigent:nov07-pr.
Authority enforcement [partial] PR actions need edit access to the session and run as the host's gh account; github_policy denies destructive operations and force pushes by default. Sources: omnigent:nov21-webui-github, omnigent:nov21-policies.
Compact PR list exposes checks/merge state and inline review discussion.
Scope
Surfaces: desktop, web, self-hosted · Version: v0.21.0; current docs checked 2026-11-07 · Plan: Not specified · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
generally-available
Delivery
native
Checked
2026-11-07 (15 days old)
Agent cross-check
Not separately cross-checked
Not established by this source pass: grouping, risk_size_filters, artifacts, request_changes, exact_revision_approval, dependency_queues, stale_approval_invalidation, check_deduplication.
12 atomic answers
Review queue [supported] Pull Requests tab lists the session's linked PRs with state and check counts. Sources: omnigent:nov07-pr, omnigent:nov07-web.
Hosted extensions [unknown] Omnigent on Databricks (Beta) is a managed service operated by Databricks. Whether it adds closed components or runs unmodified Apache-licensed code is not stated in the docs read 2026-11-21; it is a hosted service, so its components cannot be inspected.
Source visibility [supported] Public application source. Sources: omnigent:nov07-license.
Hosted parity [partial] The managed Omnigent on Databricks (Beta) does not expose custom YAML policies, bring-your-own provider API keys or custom egress controls, which the self-managed server supports. Sources: omnigent:nov21-deploy-overview.
Artifacts [unknown] Artifacts: no explicit product-scoped answer found in the checked first-party material.
Retention [partial] Archived sessions can be marked expired after 7, 30, 60 or 90 days, a per-browser setting at v0.23.0, and deleted with a Delete expired button. The blog's automatic cleanup was not found in code. Sources: omnigent:nov21-cost-blog, omnigent:nov21-src-archived-retention.
Deletion [unknown] Deletion: no explicit product-scoped answer found in the checked first-party material.
Training [unknown] Training: no explicit product-scoped answer found in the checked first-party material.
Residency [unknown] Residency: no explicit product-scoped answer found in the checked first-party material.
Subprocessors [unknown] Subprocessors: no explicit product-scoped answer found in the checked first-party material.
Consent prompt [unknown] Consent prompt: no explicit product-scoped answer found in the checked first-party material.
Provider separation [partial] Model traffic goes to the provider behind each configured credential (API key, subscription CLI, gateway or Databricks); Omnigent documents no provider data policies of its own. Sources: omnigent:nov21-models.
Omnigent's open-source self-hosted distribution is Apache-2.0 with no paid plan or price published; you bring your own vendor API key or Claude/ChatGPT subscription, billed by that provider. Billing for the vendor-managed Omnigent on Databricks (Beta) is not stated in any checked source.
Scope
Surfaces: desktop, web, self-hosted · Version: Retained findings from first-party docs and repository read 2026-11-10; omnigent.ai docs read 2026-11-21; Omnigent on Databricks (Beta) docs (updated 2026-09-28) read 2026-11-21. No installed release tested. · Plan: Open-source self-hosted distribution; billing for Omnigent on Databricks (Beta) is unknown. · OS: Not specified · Execution modes: Not specified · Region: Not specified · Harness: Not specified
Availability
unknown
Delivery
unknown
Checked
2026-11-10 (12 days old)
Agent cross-check
2026-11-22
First-party pricing pages, terms and licences read 2026-11-10, omnigent.ai docs and the Databricks Omnigent page read 2026-11-21; no purchase, billing or authenticated account test.
Pricing findings cover the open-source self-hosted distribution. Billing for Omnigent on Databricks (Beta) is not stated in any checked source.
12 atomic answers
Currency [unknown] No first-party source states currency for a paid plan. Checked: omnigent.ai, omnigent.ai/quickstart/coding-agent, omnigent-ai/omnigent README and LICENSE. No pricing page or paid plan was found, but that does not prove none exists.
Interval [unknown] No first-party source states interval for a paid plan. Checked: omnigent.ai, omnigent.ai/quickstart/coding-agent, omnigent-ai/omnigent README and LICENSE. No pricing page or paid plan was found, but that does not prove none exists.
Seats [unknown] No first-party source states seats for a paid plan. Checked: omnigent.ai, omnigent.ai/quickstart/coding-agent, omnigent-ai/omnigent README and LICENSE. No pricing page or paid plan was found, but that does not prove none exists.
Seat minimum [unknown] No first-party source states seat minimum for a paid plan. Checked: omnigent.ai, omnigent.ai/quickstart/coding-agent, omnigent-ai/omnigent README and LICENSE. No pricing page or paid plan was found, but that does not prove none exists.
Workspace fees [unknown] No first-party source states workspace fees for a paid plan. Checked: omnigent.ai, omnigent.ai/quickstart/coding-agent, omnigent-ai/omnigent README and LICENSE. No pricing page or paid plan was found, but that does not prove none exists.
Included usage [unknown] No first-party source states included usage for a paid plan. Checked: omnigent.ai, omnigent.ai/quickstart/coding-agent, omnigent-ai/omnigent README and LICENSE. No pricing page or paid plan was found, but that does not prove none exists.
Credits [unknown] No first-party source states credits for a paid plan. Checked: omnigent.ai, omnigent.ai/quickstart/coding-agent, omnigent-ai/omnigent README and LICENSE. No pricing page or paid plan was found, but that does not prove none exists.
Trial [unknown] No first-party source states trial for a paid plan. Checked: omnigent.ai, omnigent.ai/quickstart/coding-agent, omnigent-ai/omnigent README and LICENSE. No pricing page or paid plan was found, but that does not prove none exists.
Overages [unknown] No first-party source states overages for a paid plan. Checked: omnigent.ai, omnigent.ai/quickstart/coding-agent, omnigent-ai/omnigent README and LICENSE. No pricing page or paid plan was found, but that does not prove none exists.
Taxes [unknown] No first-party source states taxes for a paid plan. Checked: omnigent.ai, omnigent.ai/quickstart/coding-agent, omnigent-ai/omnigent README and LICENSE. No pricing page or paid plan was found, but that does not prove none exists.
Model cost [partial] Self-hosted: models are reached through your own first-party API key or Claude/ChatGPT subscription, so the provider bills usage. Databricks Sandbox hosts in Omnigent on Databricks (Beta) route model access through Unity Gateway and cannot use your own keys; that billing is not established. Sources: omnigent:readme-nov10, omnigent:ade157-databricks-omnigent.
Compute cost [partial] Self-hosted: remote compute is billed by the operator's sandbox provider (Databricks, Modal, Daytona and others) or hosting platform, local runners use the user's machine, and the open-source release adds no compute fee. Compute billing for Databricks Sandbox hosts in Omnigent on Databricks (Beta) is not established. Sources: omnigent:nov21-sandbox-host, omnigent:nov21-deploy-overview, omnigent:ade157-databricks-omnigent.
Access restrictions [supported] Public open-source install with no signup; the managed Omnigent on Databricks is a Beta preview enabled in workspace settings. Sources: omnigent:nov21-faq, omnigent:nov21-deploy-overview.
Status history [unknown] Status history: no explicit product-scoped answer found in the checked first-party material.
Deprecation [supported] Release notes carry a Breaking changes section, for example dropping OpenCode 1.x and macOS 12 in v0.21.0. Sources: omnigent:nov21-gh-v0210.
Export options [supported] Sessions export as JSONL from the web UI or CLI and import back. Sources: omnigent:nov21-webui-sessions.
macOS/Linux/Windows desktop shells serve full UI. Native notifications, dock badge and embedded browser. On macOS and Linux (not Windows), SSH connection starts a remote background host, then disconnects SSH; host keeps outbound server connection.
Native iOS/Android shells and full mobile web UI connect to reachable server. iOS notifications foreground-only; org-preset server URLs available on Android but still future on iOS.
Sessions persist across server restarts; projects group sessions; worktree launch options, diffs/files/inline comments and JSONL session export/import exist. Projects are grouping, not a native issue tracker.
Claude Pro/Max and ChatGPT subscriptions via official CLIs, API keys and compatible gateways. YAML agents include MCP/local function/subagent tools. Policy stacks at session/agent/server scope; caps and approval gates.
At v0.23.0 the VS Code extension (omnigent-vscode 0.1.0, publisher databricks) discovers a running local server and opens its web UI in an editor-beside iframe pane through one command, Omnigent: Open. Its README says sessions, diffs, send-selection and remote rendering are out of scope, and non-localhost server URLs are not supported. Marketplace and Open VSX releases go through a separate Databricks release repository; publication itself was not verified.
Sessions persist across browser refreshes and server restarts; Clone session forks to another agent. Export downloads <session-id>.jsonl (session metadata plus every committed item), the same format as omnigent session export and importable with omnigent session import; a failure shows an Export failed toast. Projects group sessions in non-nesting sidebar folders with per-project defaults; deleting a project archives its sessions. The composer can name a worktree branch and base, and Settings > Git can always use a random worktree. Cmd/Ctrl+/ opens a shortcuts overlay; Cmd+Opt+S (Ctrl+Alt+S) opens a name-only fuzzy session switcher and Cmd+K a palette that searches sessions. The Files panel can open a file in VS Code or Cursor. The web UI is an installable PWA.
Kind
documented
Locator
Session management; Export a session; Projects; Worktree branches; Keyboard shortcuts; Find a session by name; Install as an app; Open in VS Code or Cursor
Claude Code/Codex fully supported; Cursor/OpenCode/Pi maintained; Grok via ACP. Direct and native-TUI modes differ; plain model providers do not establish harness support.
Shipped Polly orchestrator breaks a task into sub-tasks delegated to different agents (Claude Code and Codex in the example) in parallel Git worktrees, with cross-vendor review of each PR. Agents panel shows live sub-agent status, conversations, files and terminals.
Names the harnesses Omnigent orchestrates: Claude Code, Codex, Cursor, OpenCode, Goose, Antigravity, GitHub Copilot, Devin, Kiro, Qwen Code, Kimi Code, Grok Build, Jcode, Hermes, Pi and OpenAI Agents, plus custom agents defined in YAML.
The shipped Custom ACP agents wizard prints Gemini CLI (gemini --experimental-acp) as the first example command, with Qwen Code, Goose and Claude Code.
Searched aider as a whole word across the Python package, docs and README at v0.22.0: only a code comment about other tools; no Aider harness plugin or adapter exists.
Credentials are API keys (Anthropic, OpenAI, OpenRouter, Groq, DeepSeek, xAI, Mistral, Together AI, Fireworks AI), Claude Pro/Max or ChatGPT Plus/Pro subscriptions through the official CLIs, any OpenAI- or Anthropic-compatible gateway or local server (LiteLLM, Azure OpenAI, Ollama, LM Studio, vLLM and others), or a Databricks workspace. Models absent from Omnigent's pricing catalog show no cost unless custom pricing is declared.
Kind
documented
Locator
Credential types; API key providers; Subscription; Gateway; pricing note
Omnigent is open source (Apache 2.0) and runs on your own machine and models; users bring an API key, a Claude or ChatGPT plan, a compatible gateway or a Databricks workspace, and Databricks is not required. Releases ship often and can include breaking changes, so users should pin a version and read release notes.
Kind
documented
Locator
Which models can I use?; Do I need Databricks?; Is it ready for production?
The omnigent-client package (or the full omnigent package) provides the Python omnigent_client SDK, whose version should match the server. Sessions start from omni run, the REST API or the SDK; local single-user servers need no per-request credentials. Agent tools include sys_session_create (child session), sys_session_send (follow-up to a direct child, or reuse of a named sub-agent), sys_session_message (other sessions of the same user), sys_session_list, sys_session_get_info and sys_session_get_history (read another session's recent messages and tool results).
Kind
documented
Locator
Before you start; Start a session; Let an agent manage other sessions (tool table)
Native TUI harnesses run the vendor CLI in tmux on the host and need the CLI installed and signed in; readiness problems block new sessions on that harness.
Kind
documented
Locator
Support levels; Execution modes; Native TUI harnesses need tmux
Coding-agent sessions have a Pull Requests tab listing PRs found from the agent's gh and GitHub MCP actions or linked manually. Rows show passed, pending and failed CI check counts (hover lists the checks) and refresh about every 30 seconds; the Summary tab shows reviewers with review states, CI checks and the description. For GitHub PRs users can edit title and description, labels, request or withdraw reviewers, post comments and replies, and merge after confirming a method; these need edit access to the session and run as the host's gh account. While the host is offline, rows show stored title and status and live checks and merge return when it reconnects. GitLab merge requests (via glab or a token) and Azure DevOps pull requests are read-only. On a phone the tab opens in a drawer.
Kind
documented
Locator
GitHub; Tracked pull requests; Summary tab; When the host is offline; GitLab merge requests
Recurring scheduled agent sessions use persistent task definitions, a server scheduler and a REST API. Minimum interval is ten minutes on connected hosts or continued sessions and one hour for new sandbox sessions. Overlapping fires of a task are skipped; no automatic retry or missed-fire replay.
Declarative policies allow/deny/ask actions; max tool calls, cost caps and thresholds. OS sandbox uses Linux bwrap/macOS seatbelt; plain sessions are not sandboxed by default. Generic agent-start DENY/ASK does not block launch.
Automations run a saved prompt on a recurrence on a connected host or a fresh managed sandbox. Trigger templates for Jira, Slack, GitHub, Linear and PagerDuty start one session per newly matching item (at most 10 per run) and can post the link back; they poll and do not receive webhooks, every 10 minutes by default with automations v2 or hourly otherwise, and need the service's MCP server or the gh CLI where the task runs. A failed run marks the task card and, if no session started, appears in the Inbox as Automation failed with the reason; dismissing it does not rerun the task.
Kind
documented
Locator
Scheduled tasks; Starting from a template; Failed runs; Trigger templates; PagerDuty; Change events; One task session for every trigger match
v0.21.0 lets local sessions open localhost apps in the desktop browser pane and the Files panel open files in VS Code or Cursor. Breaking changes: omnigent opencode requires OpenCode 2.0 (1.x dropped) and the desktop app requires macOS 13. If a host restarts while an agent works, the session recovers: the interrupted reply is marked interrupted and the next message runs without a manual resume.
At v0.23.0 the only webhook-style routes are POST harness callbacks in routes_hooks.py, each marked "Internal harness callback webhook — hidden from the public API reference". The search found no user-configurable outgoing or incoming webhook in the server, web UI or docs; docs/AUTOMATIONS.md says trigger templates are polling triggers, not webhooks.
Kind
source-inspected
Locator
register_hooks_routes POST handlers (lines 262-1461); search for "webhook" across .py, .ts, .tsx and .md files; docs/AUTOMATIONS.md
A cloud sandbox host moves the runner from the laptop to a remote container on Databricks Sandbox, Modal, Daytona, Blaxel, Islo, Gensee, NVIDIA OpenShell, Boxlite or AWS Lambda MicroVMs; it keeps running when the laptop closes, while a local runner stops. The server is configured once with the operator's provider credentials and anyone on the team can launch sandboxes from the web UI alongside local hosts. Lifetimes follow each provider. Third parties can add sandbox providers as plugins.
Kind
documented
Locator
Introduction and provider comparison; Server-managed; How long a sandbox stays warm; Run automations in sandboxes; Build a provider plugin
Community harnesses are separate Python packages registered under the omnigent.community.harness entry point and appear in agent YAML, --harness and the web UI picker; native-TUI harnesses are not pluggable yet.
At v0.23.0 the web UI is a React 18 and Vite app; the desktop app is Electron 45.2.1; the iOS app (Swift, Omnigent.xcodeproj) and Android app (Kotlin) are WebView shells around the server-served UI.
Users click a passage in a file or the agent's output and leave an inline comment; the comments panel separates Open and Sent comments and Address All sends every open comment to the agent in one message for revision, for example when reviewing a plan before execution. The Inbox lists pending approval requests to answer in place and sessions with unread agent output, filtered by All, Unread and Awaiting response; users with edit access can reply from a card.
Kind
documented
Locator
Inline comments and addressing; Inbox; Reply from the Inbox
Workspace content search, on by default at v0.23.0, searches file contents through the session runner with ripgrep (literal, whole-word or regex queries up to 4,096 characters); the web client hides it on mobile viewports.
Kind
source-inspected
Locator
omnigent/server/feature_flags.py lines 240-246; omnigent/server/routes/sessions/routes_resources.py line 2760; web/src/lib/contentSearch.ts; web/src/lib/capabilities.ts line 60
The scoped /tasks interface persists scheduled trigger definitions and run history; ordinary agent work is stored as sessions grouped into projects. Inspection of native stores, routes and Tasks UI found no independent backlog/to-do/issue work-item CRUD with open/in-progress/done states. Scheduled job state is not issue state.
At v0.23.0 the archived-session retention choice (Never, 7, 30, 60 or 90 days) is stored in the browser's localStorage. Settings labels it Mark as expired after and shows a Delete expired button that deletes the user's own expired archived sessions when clicked. A bounded search of omnigent/server for archived-session delete, purge, expire or cleanup found no server-side automatic deletion, which contradicts the blog's auto-cleanup wording.
Policies can ASK (route to human approval) or DENY. detect_loop ASKs when an identical tool call repeats 3 times in a 10-call window; detect_thrashing fires on 5 consecutive errors or an 80% error rate over 10 results, with ASK or DENY. spawn_bounds limits sub-agent dispatches per turn (default 5). github_policy denies destructive GitHub operations and force pushes by default. ask_on_add_policy and ask_on_scheduled_task_change are appended to every session and cannot be removed. subagent_cost_budget gates a child subtree's LLM spend; user_daily_cost_budget and user_period_cost_budget cap a user's spend across sessions.
Owners share a session with Read or Edit access; Edit collaborators send messages that execute on the owner's host through the same policy gates. Changing permission mode or launch args, transferring terminals, browsing absolute host paths and the raw environment shell stay owner-only and are enforced by the server (PATCH /v1/sessions and a 403 on the shell endpoint). Changing the session's agent stays with the owner. Admins can restrict sharing to read-only or turn it off. A fork copies history and can land on a fresh sandbox that clones the source's recorded repositories.
Kind
documented
Locator
Co-drive; What each permission level can do; Only the owner can change the agent; Fork; Fork onto a cloud sandbox
The server database is the scheduler's source of truth and re-arms active tasks on start. A connected_host task runs on a pinned or the most recently active online host; if none is live the occurrence is recorded as a failed run (host_offline). A managed_sandbox task provisions a fresh server-managed sandbox for each firing and requires a configured sandbox provider. Missed fires are not replayed (no backfill after downtime) and an occurrence that arrives while the previous run is in flight is dropped.
Kind
documented
Locator
Where a task runs (connected_host, managed_sandbox); How a firing runs: scheduling guarantees
omni claude or omni codex launches the agent in the terminal and a web UI at localhost:6767; terminal, browser and desktop app are three views of the same session and typing in one updates the others.
In Slack, replies stream into the thread and tool approvals appear as Approve / Deny cards; the web UI can answer the same request and the first answer wins.
The scoped server registers the microsandbox provider and constructs its launcher from the server configuration. Runtime providers and credentials belong to the user-operated server, rather than an Omnigent-hosted execution service.
Kind
source-inspected
Locator
microsandbox provider configuration and launcher factory, lines 1916-1925; onboarding/sandboxes/microsandbox.py
The server persists sessions in Postgres or SQLite and can be self-hosted with Docker Compose, Railway, Render, Fly.io, Hugging Face Spaces, Databricks Apps or Kubernetes. Omnigent on Databricks (Beta) is a fully managed service in which Databricks operates the server; the self-managed path is for regions without it or for custom YAML policies, bring-your-own provider API keys and custom egress controls the managed service does not expose. UIs talk to the server, never to the runner, and runners on a laptop stop with it while sandbox hosts keep running.
Omnigent on Databricks is in Beta. Databricks provides a fully managed version: a Databricks-operated Omnigent server using the workspace identity provider, model access through Foundation Model APIs and Unity Gateway, and Databricks Sandboxes in select AWS regions. It needs the Omnigent workspace preview, a Unity Gateway region and, for sandbox hosts, the Sandbox preview. Custom Python policies are not supported and Sandbox hosts cannot use your own model API keys. The page states no price.
Kind
documented
Locator
Beta notice; managed version list; Availability; Limitations; Last updated
The optional microsandbox provider boots each Omnigent host in its own hardware-isolated microVM using libkrun. It is fully local/self-hosted, supports Apple Silicon macOS and Linux with KVM, and uses a beta pinned runtime. Server-managed sessions and CLI sandbox create are documented.
Kind
documented
Locator
Provider overview; prerequisites; server-managed and CLI bootstrap
Omnibox wraps an agent in an OS sandbox (bubblewrap and seccomp on Linux, Seatbelt on macOS). With egress_rules, HTTP(S) leaves through a default-deny proxy with an allow-list of methods, hosts and paths, and private IPs and metadata endpoints are blocked. The agent holds a placeholder token and the proxy swaps in the real credential for allowed requests, so the secret never appears in logs, transcripts or model context.
The /usage page shows a daily cost timeline, cost by harness, cost by model and a sortable session table; multi-model and multi-harness sessions show every model and harness used. The post says archived sessions are auto-cleaned on a schedule you set (7, 30, 60 or 90 days) under Settings > Archived sessions.
Session PR associations persist server-side; description, checks, comments, changed files and diffs stay on and are served by the host provider integration. Incomplete data is flagged; linked request summaries remain when host offline; head/base revisions are validated before diffs are returned.
Kind
documented
Locator
Presentation; server storage; inline conversations
Omnigent budget policies enforce recorded cost at supported turn or tool-call gates; unpriced tokens prompt to continue untracked. They are not a provider billing limit or an infrastructure budget and do not control clients outside managed sessions. Page last checked against v0.19.0.
Kind
documented
Locator
How do I cap coding-agent spending in Omnigent?; checkpoints and hard caps
Admins set a server-wide session-sharing policy in Settings > Sharing or with OMNIGENT_SHARING_MODE, which can limit sharing to read-only or disable it.
Invite-only built-in accounts or OIDC/header SSO. Admin manages membership/policies/sharing; owner/Edit/Read permissions scope sessions. First admin setup must be completed by operator.
Phones can use native iOS and Android shells or the full touch-optimized web UI in any phone browser for chat, sub-agents, files, code changes and inline comments.
Sharing with Edit access lets a teammate direct the existing agent in its environment on the same reachable server; a session link is not a Git clone and need not carry uncommitted files. Page last checked against v0.19.0.
Kind
documented
Locator
How do I share a coding-agent session in Omnigent?; Should I share a session or send a Git branch?
Anonymous server usage telemetry defaults on; opt out with OMNIGENT_ANALYTICS=0, DO_NOT_TRACK=1 or telemetry:false. Prompts/files/credentials excluded according to vendor.
Says Omnigent works with any model, using a first-party vendor API key or an existing Claude Pro/Max or ChatGPT subscription. It also offers cost-budget policies.
The only match was a web_read tool comment about pages behind paywalls; no Omnigent payment, card or subscription flow was found in the self-hosted code.
Kind
source-inspected
Locator
omnigent/ and web/src searched for stripe, paywall, billing_portal, checkout_session