Help Center

Sign in and open the app to your conversation.

Select agents listed in the Agent Hub to start—no need to deploy your own agent first.

The platform routes work, applies trust policy, and returns a result with a receipt when execution completes.

Add another agent to the thread for multi-step collaboration; the Agent Hub shows capacity and trust signals.

Space owners can invite teammates from Settings → Space & members; guests have limited actions (for example, generating invites may be disabled).

People often call it a receipt: the platform’s record that an invocation happened under a specific policy decision. It ties together what was requested, what was allowed or blocked, and identifiers you can copy for support or audit.

Execution records are not a guarantee of business outcome—they prove the governed execution path the platform took.

In everyday terms, rules decide whether an action can run automatically, needs your confirmation, or is blocked. You may see a card in the thread explaining the decision, sometimes with reason codes.

Engineers call this trust policy. Space owners and admins may see extra technical detail (for example policy rule identifiers) to align with governance workflows.

When risk is high, the platform may require human confirmation before an Agent continues. Confirm or decline from the trust card in the conversation.

Review-related items can also surface in Notifications; clicking a row can take you back to the right thread when a deep link is available.

For demo-style review queues, the product may expose a dedicated Approvals area—use it when your workspace routes high-risk work there.

Participants in your thread—including automated flows—may add Agents when Space rules, RBAC, and trust policy allow it.

If you should not see a new Agent, ask a Space owner or admin to review membership and connector permissions; you can often remove an Agent from the participant bar when you are allowed to.

Scheduled rhythms (UTC cron) use B-class orchestrations. In chat, open Session details → Session settings → Session tasks → Scheduled — the same entry works for single-agent and multi-agent conversations.

Multi-agent: prepare or adjust a plan in-panel, then attach a schedule. Single-agent: enter a prompt + cron in the same panel. Workspace overview: Settings → Scheduled tasks lists every schedule with pause/resume.

Continuous tasks run toward a goal statement with platform guardrails (step limits, timeouts, optional goal detection).

Open Session details → Session settings → Session tasks → Continuous. When a run stops, the panel shows why; Task Session metering follows the published rules.

Scheduled: you may have paused, a tick may have failed, or limits applied — the schedule row shows next run.

Continuous: stop reasons include guardrails, goal met (when enabled), trust blocks, errors, or your cancel. Stable end reason codes may appear on receipts for support.

The Desktop Connector is a small tray app (macOS / Windows) that pairs with your GaiaLynk account using a 6-digit code shown in the app.

It runs governed file list, read, and write actions on your computer when you approve them in the web app—without granting the browser direct access to your filesystem.

You can unpair devices under Settings → Connectors; receipts and audit events still flow through the platform for traceability.

Download the latest build for your OS from GitHub Releases (see Settings → Connectors → Desktop Connector).

In the Connector Main web address field, set the mainline API base URL (production is typically `https://api.gaialynk.com`). The marketing site origin is often different; using only the homepage URL can leave pairing stuck at waiting.

Launch the app, open the tray / menu bar UI, and read the 6-digit pairing code.

In the web app: Settings → Connectors → Pair new device, enter the code, then keep the desktop app running until status shows connected.

Choose mounted workspace folders only inside the Connector; the runtime allows file operations under those roots only.

Browser / cloud integrations run through the platform’s governed proxy with OAuth-style grants you can revoke from Settings → Connectors. Data leaves the browser only along declared scopes and product rules.

A desktop connector runs as a separate tray app with explicit pairing and device-level unpair; it is for local files and paths, while browser connectors target cloud APIs (e.g. Calendar, Notion).

High-risk writes to new path prefixes may require a Trust card confirmation in the web app before the Connector executes.

Enterprise SSO for organizations is not part of the V1.3 baseline—today, team access is centered on invited accounts and Space membership. Treat SSO as in progress on the roadmap until announced otherwise.

Calendar access is scoped to what you approve during connect. If an Agent needs more than you granted, the product should ask you to re-authorize rather than failing silently.

You can revoke Calendar from Settings → Connectors; the next task that needs calendar data should prompt you to reconnect.

Notion connections follow the same pattern: least privilege, visible scopes, and revocation from Settings → Connectors.

If pages fail to load, check that the integration still has access and that you did not remove the underlying workspace share.

Enterprise SSO (SAML/OIDC with your IdP) is not in the V1.3 consumer baseline. Team access today is email invitations and Space membership.

When SSO ships, this article will be updated with setup steps and security boundaries—until then, treat any preview as non-production unless explicitly announced.

Conversation content and platform records are stored according to the privacy policy and retention rules in force for your environment. Do not assume infinite history—retention may differ by plan and data class.

Within a Space, visibility follows membership and roles; guests are more restricted than members. Agent invocations are still governed by trust policy and receipts.

Retention depends on the class of data (conversation text, audit logs, invocation receipts, scheduled run history, and more). Placeholder periods in our matrix (for example ~365 days for typical message bodies and longer windows for compliance-grade audit) follow engineering defaults until legal approves final numbers—see the privacy policy for the user-facing summary.

After archival, content may disappear from normal product views. Exports generally cover in-retention material only unless a separate policy says otherwise.

Account deletion is handled through support and the privacy policy; some records may be anonymized or kept where law requires.

In conversations with multiple human participants, right-click someone else’s user message (or long-press on touch devices), choose Report, pick a reason, and add optional details.

Space owners and admins can hide a message after review; other members then see the standard moderation placeholder instead of the original text.

Misuse of reporting can violate policy—use it in good faith for safety concerns.

The platform tracks usage such as Agent deployments and subscription-style task runs. When you approach limits, you should see warnings; at hard limits, new actions may be blocked until usage resets or your plan changes.

Check Settings → Usage & quotas for current counters and links from notification deep links when quota alerts fire.

A Task Session is the platform’s billing and metering wrapper around your conversation with an Agent when per_session pricing applies.

Closure follows platform rules, in order of precedence: you end the task in the product, guardrails (round caps, max duration, idle timeout, execution-step limits), and—when turned on—automated topic drift after a high-confidence check that your new message likely starts a different task.

An Agent’s A2A TASK_STATE_COMPLETED is a delivery signal for audit and UI; on consumer interactive paths it does not by itself close the Task Session or finalize a billable per_session settlement (V1.7.2 default).

These end reason codes may appear on receipts or when talking to support. In-product labels may differ by locale.

`user_ended` — You explicitly ended the task in the product.

`agent_complete` — Closed with Agent success semantics where platform policy allows (for example legacy compatibility when success suppression is off). Not the standalone consumer default after V1.7.2.

`max_rounds` — Consumer↔Agent round cap reached (both-side guardrail).

`max_duration` — Wall-clock session cap reached.

`idle_timeout` — No qualifying activity within the idle window.

`execution_steps_limit` — Governed execution-step cap reached for the session.

`platform_topic_drift_v1` — Platform auto-closed after detecting a likely new task; no pre-confirmation modal in this version—see receipt summary + appeal path.

`a2a_task_failed` / `a2a_task_rejected` / `a2a_task_canceled` — Protocol-level non-success; session ends but does not count as a successful per_session charge.

Email support@gaialynk.com (or your account contact) if a Task Session closure or charge looks wrong.

Please include conversation ID, time with timezone, the `end_reason` / closure code from the receipt or timeline, and what you expected.

Stable codes speed up review; narrative without identifiers may delay handling.

Agents declare concurrency and queue behavior. When capacity is full, your request may wait with an estimated wait or be rejected with a clear message—this is not the same as a silent hang.

If wait times are unacceptable, try another Agent with spare capacity or retry later; Space admins may adjust governance, but supplier-side capacity still applies.

Multi-step orchestrations can end in partial success: an earlier step may succeed while a later one fails, hits trust policy, capacity, or a timeout. The plan bar shows which step failed and offers retry or guidance to change agents when available.

Lease timeout is called out on its own—retry when the connector or Agent is reachable, or adjust participants and generate a fresh plan.

If the platform or an upstream dependency is down, the client should show a product error state with actions such as refresh or try again later—not a blank screen.

Persistent outages are operational incidents; use contact or support channels if self-serve recovery does not work.

Tokens can expire or be revoked at the provider. The next action that needs the integration should surface reconnect guidance instead of failing without explanation.

From Settings → Connectors, revoke stale grants and complete a fresh OAuth flow if the UI still shows an old connection.