Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.
JavaScript 3d-generationai-agentsanimationavatardigital-humangenerative-aiglbgltfimage-to-3dllmmcpmcp-servermodel-context-protocolsolanatext-to-3dthreejsweb-componentswebglwebxrx402
// GET /api/cron/reflect-sweep — scheduled memory-consolidation pass.
//
// Walks agents with recent activity (new memories or actions) that haven't
// reflected within the debounce window and runs a real reflection pass for each
// (api/_lib/reflection.js). This is the "while you were away" engine: the user
// comes back to find their agent has genuinely consolidated what it saw.
//
// Cost discipline (no silent caps):
// • A cheap SQL pre-filter skips agents with no recent activity or a run in the
// last 30 minutes, so the engine isn't invoked for idle agents.
// • The engine itself re-checks the per-agent daily cap + debounce + minimum
// new-signal threshold and records every skip in agent_reflection_runs.
// • A wall-clock budget bounds the batch; whatever it couldn't reach is logged
// and picked up on the next tick — never silently dropped.
import { sql } from '../_lib/db.js';
import { cors, json, method, wrapCron } from '../_lib/http.js';
import { runReflection } from '../_lib/reflection.js';
import { requireCron } from '../_lib/cron-auth.js';
export const maxDuration = 120;
// How many agents to attempt per tick, and the wall-clock budget. The cron runs
// every 30 min, so a backlog drains over a few ticks rather than in one long run.
const BATCH_LIMIT = Number(process.env.REFLECT_SWEEP_BATCH) || 12;
const TIME_BUDGET_MS = 100_000;
export default wrapCron(async (req, res) => {
if (cors(req, res, { methods: 'GET,OPTIONS' })) return;
if (!method(req, res, ['GET'])) return;
if (!requireCron(req, res)) return;
const startedAt = Date.now();
// Pre-filter: agents with activity in the last 36h and no reflection run in
// the last 30 minutes (the engine's debounce is authoritative; this just
// avoids waking it for agents that would only be told to wait).
const candidates = await sql`
WITH active AS (
SELECT agent_id FROM agent_memories WHERE created_at > now() - interval '36 hours'
UNION
SELECT agent_id FROM agent_actions WHERE created_at > now() - interval '36 hours'
)
SELECT a.id, a.name, a.description, a.user_id
FROM agent_identities a
JOIN active ac ON ac.agent_id = a.id
WHERE a.deleted_at IS NULL
AND NOT EXISTS (
SELECT 1 FROM agent_reflection_runs r
WHERE r.agent_id = a.id AND r.created_at > now() - interval '30 minutes'
)
ORDER BY a.id
LIMIT ${BATCH_LIMIT}
`;
const summary = { eligible: candidates.length, processed: 0, dreamsCreated: 0, skipped: 0, errors: 0, deferred: 0 };
for (const agent of candidates) {
if (Date.now() - startedAt > TIME_BUDGET_MS) {
summary.deferred = candidates.length - summary.processed;
console.log(`[reflect-sweep] time budget reached; deferring ${summary.deferred} agents to next tick`);
break;
}
summary.processed++;
try {
const result = await runReflection({
agentId: agent.id,
userId: agent.user_id,
trigger: 'cron',
agent: { name: agent.name, description: agent.description },
});
if (result.status === 'ok') summary.dreamsCreated += result.created.length;
else if (result.status === 'skipped') summary.skipped++;
else {
summary.errors++;
// runReflection reports engine failures by return value, not by throwing
// (it records the run either way). Without this the response says
// "errors: 1" and nothing anywhere says which agent or why.
summary.lastError = result.reason || 'reflection failed';
console.error(`[reflect-sweep] agent=${agent.id} reflection error:`, summary.lastError);
}
} catch (err) {
summary.errors++;
console.error(`[reflect-sweep] agent=${agent.id} failed:`, err?.message);
}
}
summary.elapsedMs = Date.now() - startedAt;
return json(res, 200, summary);
}, { maxDurationMs: maxDuration * 1000 });