Real-time short-form video intelligence across TikTok, YouTube Shorts, and Instagram Reels, powered by Virlo Content Research Agents.
The official Virlo plugin for Vellum assistants. Gives your always-on personal AI real-time short-form social intelligence: ask "what's going viral in my niche right now, and why?" and get back Virlo's weighted virality score, rising-creator outliers, winning content formats, and AI trend analysis across TikTok, YouTube Shorts, and Instagram Reels.
# 1. Hatch a Vellum assistant
vellum hatch --name my-assistant --remote docker -d
# 2. Store your Virlo API key (get one at https://dev.virlo.ai/dashboard)
vellum exec my-assistant -- assistant credentials set "virlo_tkn_<your_key>" \
--service virlo --field api_key
# 3. Install the Virlo plugin (triggers the init hook)
vellum exec my-assistant -- assistant plugins install \
https://github.com/Virlo-AI/virlo-integrations/tree/main/integrations/vellum --name virlo
# 4. Start a conversation
vellum message my-assistant "what's going viral in fitness on TikTok right now?"
The assistant will resolve the API key from the credential store at runtime, kick off a Content Research Agent (~15-20 min), and surface results when ready. It never blocks you synchronously.
Built on Virlo's forward-facing surfaces:
/v1/agents) - the unified research engine. One resource, two modes:
is_recurring: false) - niche search: "what's viral in London TikTok fitness right now?"is_recurring: true) - monitoring: "keep watching this niche and tell me weekly."emerging feed for "what's about to take off."integrations/vellum/
package.json Plugin manifest (name, version, peerDependencies)
mcp.json MCP server connection config (for future MCP support)
hooks/
init.ts Logs plugin registration (no config/env state)
shutdown.ts No-op (no state to clean up)
skills/
virlo/
SKILL.md The assistant skill: how to use Virlo's API
scripts/ Runnable TypeScript scripts for each flow
virlo-client.ts Shared helper (credential resolution, fetch, polling)
whats-viral.ts One-shot niche search flow
recurring-monitor.ts Recurring monitor creation
creator-deep-dive.ts Satellite creator lookup + video outlier
references/ Worked flow documentation + golden prompts
agent-playbook.md How to interpret results (weighted score, formats)
golden-prompts.md Acceptance criteria for QA
whats-viral-in-my-niche.md
recurring-niche-monitor.md
creator-deep-dive.md
bun run.bash tool with assistant credentials reveal --service virlo --field api_key.virlo_tkn_).assistant credentials set --service virlo --field api_key "virlo_tkn_<your_key>"
Bring-your-own-key. Tiering lives in Virlo. Free vs. paid gating, credits, and rate limits are all enforced server-side by the Virlo API - the plugin never gates features, it just relays what the user's key is allowed to do. Users pay Virlo directly from their own prepaid balance.
402 when a user is out of credits - the assistant surfaces this and links to billing.Content Research Agent runs (one-shot and recurring) take ~15-20 minutes (up to 45 with Meta ads). The scripts poll until finalized: true, or the assistant can subscribe to the content_research_agent.run.completed webhook, then notify the user - it never blocks them synchronously. This is a natural fit for Vellum's always-on model.
MIT (c) Virlo