Orkas is an open-source (MIT), local-first AI team for your desktop: a Commander turns your goal into to-dos and coordinates 16 built-in specialist agents plus Claude Code, Codex, OpenCode, OpenClaw and Hermes, with projects, scheduled automations and connectors. Free app for macOS & Windows; built-in models or your own keys.
JavaScript agent-collaborationagent-harnessagent-orchestrationagent-teamagent-workflowai-assistantai-automationai-employeeai-productivityai-teamai-workflowautonomous-agentsclaude-code-guiclaude-coworkdesktop-agentlocal-firstmulti-agentmulti-agent-systemself-evolving-agentstask-automation
// ─── Model config guard ──────────────────────────────────────────────────
// Central gate that keeps LLM-dependent actions disabled until the user has
// at least one (provider, model, credential) entry. `ensureModelConfigured()`
// is the synchronous action boundary: it short-circuits with an alert and a
// Settings redirect when no model is available. There is intentionally no
// persistent page-level warning; the user only needs guidance when starting
// an action that requires a model.
// Refreshed at boot and after Settings reloads following a successful
// credential/entry transaction, OAuth entry creation, repair, or deletion.
const _guardLog = createLogger('model-guard');
function _modelGuardErrorType(error) {
return error && typeof error.name === 'string' ? error.name : 'unknown';
}
let _hasConfiguredModel = true; // optimistic — flipped to false after refresh if empty
let _modelGuardRefreshSequence = 0;
async function refreshModelGuard() {
const refreshSequence = ++_modelGuardRefreshSequence;
try {
const res = await window.orkas.invoke('auth.hasConfiguredModel');
// Only flip the flag when the IPC returned a definitive answer. A
// failed call (unknown channel on an old main process, transient
// error) should leave the UI optimistic rather than lock the user
// out of every feature. The backend runner still fails loudly on
// actual sends if no entry exists, so we don't lose correctness.
if (res && res.ok) {
const configured = !!res.configured;
await refreshModelConfigSnapshot();
if (refreshSequence !== _modelGuardRefreshSequence) return _hasConfiguredModel;
_hasConfiguredModel = configured;
} else {
_guardLog.warn('refresh ipc not-ok', { error: res && res.error });
}
} catch (e) {
_guardLog.warn('refresh failed', { error: (e && e.message) || String(e) });
}
if (refreshSequence !== _modelGuardRefreshSequence) return _hasConfiguredModel;
return _hasConfiguredModel;
}
async function refreshModelConfigSnapshot() {
try {
const res = await window.orkas.invoke('auth.listEntries');
if (res && res.ok && Array.isArray(res.entries)) {
window.dispatchEvent(new CustomEvent('orkas:model-entries-changed', {
detail: { entries: res.entries },
}));
}
} catch (err) {
_guardLog.warn('model config snapshot refresh failed', { error: (err && err.message) || String(err) });
}
}
function isModelConfigured() {
return _hasConfiguredModel;
}
/**
* Synchronous gate used by action handlers before firing an LLM-backed
* request. Returns `true` when configured (caller proceeds); `false` after
* showing an alert + bouncing the user to the settings page.
*
* `opts.silent === true` skips the alert — useful for background/auto paths
* (queued sends, polling) where we just want to quietly no-op.
*/
function ensureModelConfigured(opts = {}) {
if (_hasConfiguredModel) return true;
if (!opts.silent) {
const msg = opts.message || t('model_guard.modal');
try {
if (typeof uiAlert === 'function') uiAlert(msg);
else window.alert(msg);
} catch (_) { /* swallow — alert is best-effort */ }
if (typeof setView === 'function') {
setView('settings', null, { entryPoint: 'model_guard_blocked_action' });
}
_activateModelCredentialsTab();
}
return false;
}
function _activateModelCredentialsTab() {
const activate = () => {
if (typeof window.activateSettingsTab !== 'function') return false;
window.activateSettingsTab('credentials');
return true;
};
// Settings can be initialized lazily by setView(). Its first render uses
// the default Account tab. Wait for the owning feature when necessary,
// then repeat on the next turn to win its initial-render race.
if (!activate()) {
const load = (typeof loadRendererFeature === 'function')
? loadRendererFeature
: window.loadRendererFeature;
if (typeof load === 'function') {
Promise.resolve(load('settings')).then(activate).catch(() => {});
}
}
setTimeout(activate, 0);
}