Documentation

AgentHub Documentation

AgentHub is TokenDance's multi-agent engineering collaboration product line. It connects AgentHub Desktop, the Web workbench, Hub Server, Edge Server, and Agent Runtime adapters so that coding agents such as Claude Code, Codex, and OpenCode can work inside visible team workflows.

Current status

AgentHub is under active development. The public website and docs are live; Desktop + Local Edge is the preview-ready local path; Hub Server and Web workbench are in progress; Feishu/Lark production integration, Remote/Cloud Edge, full Web + Hub + Edge routing, and several database-backed surfaces are still in development.

Start Here

If you are new to AgentHub, choose the path that matches your role, then use the ordered list as the full tour.

ReaderRead firstOutcome
Product evaluatorProduct Status -> Comparison -> RoadmapUnderstand what is live, what is preview-only, and why AgentHub exists alongside CLI tools
Local userInstallation -> Desktop Guide -> QuickstartRun the smallest Desktop -> Local Edge -> Runtime adapter path safely
Runtime integratorAgent Profiles -> Run Lifecycle -> Adapters -> API And EventsMap a runtime into AgentHub's event, approval, diff, and artifact model
Team/workflow ownerCollaboration -> Hub And Edge -> Web Workbench -> Feishu/Lark IntegrationPlan shared sessions, Web review, Hub routing, IM entry points, and evidence
Release ownerSecurity -> Operations Runbook -> Deployment -> Release ChecklistVerify docs, visual QA, SEO, public discovery files, and safe release wording

Full recommended reading order:

  1. Concepts: understand Desktop, Web, Hub, Edge, Agent Profile, Runtime adapter, Run, and Event.
  2. Product Status: separate live, preview-ready, contract-shaped, in-progress, and in-development capabilities.
  3. Quickstart: start the smallest local flow and verify Desktop, Local Edge, and a runtime adapter can work together.
  4. Installation: prepare a workstation with Git, Go, Node.js, pnpm, Local Edge, Desktop, and a safe runtime path.
  5. Glossary: align product, runtime, identity, integration, and release terms.
  6. Workflows: map local runs, reviewable diffs, team collaboration, Feishu/Lark tasks, adapters, and releases to concrete evidence.
  7. Usage Cookbook: follow concrete recipes for read-only reviews, reviewable diffs, adapter comparison, failed-run triage, team task prep, Feishu/Lark prep, and docs releases.
  8. Desktop Guide: understand the local workbench, runtime selector, diff review, approvals, and QA evidence.
  9. Desktop UI Reference: review layout, run states, selectors, diff panels, motion, and screenshot QA.
  10. Configuration: review local, Hub/Web, runtime profile, and secret ownership before wiring real credentials.
  11. Identity And Login: confirm TokenDance ID login, callback, browser state, and authorization boundaries.
  12. Architecture: learn the Hub / Edge / Desktop / Web responsibility boundaries.
  13. Hub And Edge: check the collaboration/execution split, routing, event contract, and integration entry boundary.
  14. Web Workbench: understand Hub-backed Web surfaces, local-file boundaries, review flow, and evidence.
  15. Agent Profiles: define profile metadata, capabilities, approval policy, and adapter boundaries.
  16. Collaboration: map shared sessions, team roles, approvals, integration entries, and evidence.
  17. Run Lifecycle: follow run states, event envelopes, approval gates, artifacts, diffs, and failure codes.
  18. API And Events: review the public REST, WebSocket event, and runtime adapter contract boundaries.
  19. FAQ: answer common product, setup, identity, runtime, integration, and release questions.
  20. Design System: follow TokenDance Blue, Desktop mock, icon, motion, and visual QA rules.
  21. Deployment: use the static-export, smoke, and stale-page triage guide.

Documentation Map

AreaWhat it coversStatus
ConceptsTerms, product surfaces, delivery status, and boundariesLiving document
Product StatusLive, preview-ready, contract-shaped, in-progress, and in-development capability mapLiving document
GlossaryProduct, runtime, identity, integration, and release terminologyLiving document
WorkflowsLocal runs, diff review, collaboration, Feishu/Lark, adapter, and release workflowsLiving document
Usage CookbookPractical recipes for read-only reviews, diff review, adapter comparison, failed-run triage, team tasks, Feishu/Lark prep, and docs releasesLiving document
InstallationWorkstation setup, runtime CLI preparation, local ports, verification, and first failure triageLiving document
Desktop GuideDesktop surface map, runtime controls, diff review, approvals, theme/language evidenceLiving document
Desktop UI ReferenceLayout model, run states, selectors, runtime picker, diff panels, motion, and screenshot QALiving document
QuickstartLocal preview, configuration, and first agent taskPreview-ready
ConfigurationLocal Edge, Hub/Web, runtime profile, and secret ownershipLiving document
Identity And LoginTokenDance ID login flow, callback, browser state, authorization boundary, and UI rulesLiving document
ArchitectureHub, Edge, Desktop, Web, runtime adapters, and event flowPreview-ready
Hub And EdgeHub/Edge responsibility split, routing, event contract, and integration entriesLiving document
Web WorkbenchHub-backed Web surfaces, local-file boundary, review flow, UI states, and evidenceIn progress
Agent ProfilesProfile model, capability vocabulary, selection rules, approval policy, and adapter boundaryContract shaped
CollaborationShared sessions, review/approval flow, team roles, integration entries, and evidenceIn progress
Run LifecycleRun state machine, event envelope, approval gates, artifacts, diff metadata, and failure codesContract shaped
AdaptersClaude Code, Codex, OpenCode, and custom adapter contractsContract shaped, public SDK not stable
API And EventsHub API, Edge API, WebSocket events, adapter eventsIn development
Feishu/Lark IntegrationBot, events, cards, H5/workbench, TokenDance ID bindingIn development
SecurityTokenDance ID, Hub-local permissions, secrets, sandbox, and auditLiving document
Operations RunbookPublic-site checks, docs route release gate, visual QA, smoke, and failure triageLiving document
DeploymentRelease source, local build, static export, live smoke, and stale-page triageLiving document
TroubleshootingSetup, login, runtime, Feishu, and static-site failuresLiving document
FAQProduct, setup, identity, runtime, integration, docs, and release questionsLiving document
Release ChecklistDocs, SEO, visual QA, security, and deployment gatesLiving document
RoadmapCapability status and conservative release languageLiving document
Design SystemTokenDance Blue, component rules, Desktop mock, motion, icon, and visual QA guidanceLiving document
ComparisonPositioning against CLIs, chat UIs, and single-agent toolsLiving document
ChangelogPublic docs and product-site changesLiving document

What AgentHub Is

Most AI coding tools start as a private terminal session: one person launches the agent, watches the logs, and decides whether the result is useful. AgentHub gives that work shared state, review steps, and records that a team can inspect.

In the current design:

  • AgentHub Desktop is the local workbench. It connects to Local Edge and is suited for personal development and local execution.
  • AgentHub Web is the Hub-backed collaboration workbench for multi-device, team, IM, and remote workflows.
  • Hub Server owns identity sessions, projects, devices, IM entry points, sync, and audit.
  • Edge Server owns workspaces, run lifecycle, runtime adapters, diff, artifacts, and preview events.
  • Runtime adapters host real execution engines such as Claude Code, Codex, and OpenCode.

AgentHub is a controlled workbench for agent work. It shows who requested a task, which runtime executed it, what files changed, which approvals were requested, what record was produced, and what can be reviewed later.

Current Boundaries

You can develop and preview these areas today:

  • Desktop connects to Local Edge and can validate mock / Claude Code / Codex / OpenCode preset execution flows.
  • Web can run as a Hub-only preview surface. It does not start local CLI processes or bypass the Hub to connect directly to Local Edge.
  • Edge Server has the runtime adapter structure and local execution boundaries.
  • Hub Server has local development paths for OIDC code exchange, Hub sessions, device routing, and task routing.

Still in development:

  • Feishu/Lark production event ingress, card schema, async queues, and account-binding flow.
  • Remote/Cloud Edge, device proof, workspace allowlist, and relay/provisioning.
  • Database-backed Contacts, Docs, Tasks, Projects, and Settings surfaces.
  • Production-grade full Web + Hub + Edge routing E2E verification.
  • Public third-party Adapter SDK package and ecosystem submission flow.

When a page says "preview" or "in development", treat that wording literally. Public docs should help contributors and early users understand the intended system without implying that every surface is generally available.

Identity

TokenDance ID is the identity authority for AgentHub. AgentHub does not own a direct GitHub, Google, or Feishu social login, and Feishu OAuth must not be treated as product login. Hub Server can issue AgentHub-local sessions and enforce product permissions; low-risk Desktop/Web personalization is not a Hub API authorization boundary.

Next Step

Continue with Concepts, map your scenario in Workflows, then run the local path in Quickstart.