Skip to content

FAQ

Same heartbeat / multi-agent / group-rooms primitives. The differences are in the deployment model:

  • Multi-tenant SaaS — your team shares a tenant rather than each running their own self-host
  • EU-sovereign hosting — data stays in EU regions
  • Multi-LLM routing — Anthropic, OpenAI, Google, Mistral; the platform routes per-call so cost can be 10× lower for routine heartbeats (Haiku / 4o-mini / Flash)
  • Vision-loop for strategic agent autonomy (multi-phase ASSESS → DEBATE → PLAN → EXECUTE → REVIEW) — not just heartbeats
  • Approval queue, audit log, security gate built in for compliance use
  • Packaged industry apps (Office Assistant, Sachverständiger, Marketing War Room) ready to install

OpenClaw is the better choice for personal-productivity self-host; Linkworld for businesses that need governance, multi-LLM economics, and pre-built apps.

Yes by default. The platform’s primary deploy is EU regions; tenant databases, audit logs, agent memory, and uploaded files all sit in EU infrastructure. Cross-region replication is opt-in.

LLM provider calls go to the provider’s infrastructure (Anthropic US, OpenAI US, Google EU, Mistral EU). The platform doesn’t automatically choose EU-only — you configure your tenant’s LLM provider preference. If EU-only routing matters for your use case, choose Mistral or Google’s EU regions in the tenant settings.

Depends on the model tier you pick. Cheap-tier heartbeats running on Gemini Flash or GPT-4o-mini cost roughly €0.05 / agent / day for a typical 5–10 heartbeat agent. Premium-tier on Sonnet 4.6 is ~10× that.

The pre-filter mechanism means most ticks (≈ 80 %) skip the LLM entirely, which keeps the bill predictable. The audit log captures per-call tokens so you can see costs in detail.

Three layers of protection:

  1. Manifest declarations — apps must declare what they can do at install time. Adding a new tool requires a new app version and your re-approval.
  2. Bilateral grants — cross-app calls require both apps’ admins to opt in. The Steward can’t reach across the tenant automatically; you set up each pairing explicitly.
  3. Approval queue — destructive / sensitive actions surface as approval rows you sign off on individually.

You can run agents as fully-autonomous (skip the approval queue), fully-supervised (every action requires approval), or anywhere between (per-tool approval policy). Default is supervised — deliberately conservative.

Three brakes:

  • Auto-disable on failure — heartbeats with 5 consecutive failures auto-disable. Failing earlier shows up on the Health screen.
  • Pause an app — Workspace Control → app row → Pause stops all heartbeats and disables the agents while keeping memory intact.
  • Revoke a grant — wires & grants editor, click Revoke. New cross-app calls denied immediately.

Can the same agent run for multiple users?

Section titled “Can the same agent run for multiple users?”

Agents are tenant-scoped, not user-scoped. One tenant’s agents are isolated from another tenant’s. Within a tenant, all users share the same agent set (the agent’s memory is shared, the conversation is per-user).

If you need per-user isolation within a tenant (rare), the recommended pattern is multiple tenants — invite-based, each user runs their own. The platform supports this; your billing scales with active tenants.

Yes. As tenant admin:

  • App records: per-app export from the app’s UI (or the bulk export endpoint at /api/admin/records/export)
  • Audit log: time-range export from the Audit screen (“Export CSV / JSON”)
  • Conversations + memory: per-user export from the user’s profile menu
  • Full tenant export: contact support for a tenant-wide bundle (usually within 24 h)

All exports are JSON-Lines or CSV, machine-readable.

Default retention windows:

DataDefault
Audit log365 days
Approvals365 days
Agent memoryindefinite
Conversationsindefinite
Recordsindefinite

Configurable per tenant within legal limits. Beyond default windows, archived data lives in cold storage.

The standard offering is hosted SaaS. On-prem / private-cloud is available for enterprise deals — you run the platform in your infrastructure (EU sovereign cloud, your own AWS / Azure account, or even airgapped). Contact sales for pricing.

Note: on-prem doesn’t get automatic platform updates. You manage the deploy cadence yourself.

Email [email protected]. PGP key on the website.

For non-security bugs: GitHub issues at github.com/linkworld/linkworld.

Workspace Control → Health → “Tenant” in the page header (or URL slug). When opening support tickets, include the tenant ID + the timestamp of the issue.

Why doesn’t agent X respond when I @mention it in a room?

Section titled “Why doesn’t agent X respond when I @mention it in a room?”

Three causes:

  1. Not a member: only members hear @mentions. Add the agent from the room’s members list.
  2. App is paused: the agent’s app might be paused — check the dashboard.
  3. Agent declined: agents can decide not to respond (e.g. the prompt didn’t make them think they should chime in). This is a feature, not a bug — the platform doesn’t force chatbot- style auto-responses.

Try a more direct prompt; “@cmo: please respond” rather than @cmo in a sentence about something else.