105128931081781909095Q: API & persistent agents and limits
1) Can external platforms such as Paperclip, Hermes, Antigravity or n8n communicate with Agently through APIs, MCP or webhooks?
Can an external orchestrator create a task in Agently, assign it to Jarvis or a specialist, monitor its status and retrieve the completed output?
2. Windows desktop application planned?
3. Can multiple specialist subagents work on different tasks simultaneously, and can I control the maximum concurrency?
Are there limits on the number of subagents per task, the number of tasks per day, or the number of simultaneous workflows?
Are there queue limits, execution timeouts, rate limits or restrictions on long-running research tasks?
Are these limits determined by the plan, model provider, infrastructure capacity or a combination of these factors?
ahmadhajj
Oct 14, 2026A: Thanks for the detailed questions!
1. External platforms. Yes. Agently has an MCP server that tools like n8n, Hermes, Claude or your own orchestrator can connect to. Through it, an external platform can:
- create a task and assign it to Jarvis, a specialist or one of your custom agents, optionally with a goal and a credit budget;
- check its status, progress summary and next step;
- fetch the finished result;
- stop it.
It can also search your company's memory, use your skills and save pages. Everything goes through the same permissions, approvals and credits as work you start in the app. Outgoing webhooks on completion are coming next; for now, orchestrators check the task's status.
2. Windows. Yes, a Windows desktop app is planned.
3. Concurrency and limits.
- Sub-agents: Jarvis can run up to 5 specialist sub-agents in parallel per task, and each one keeps working until its part is done, even across restarts. Sub-agents don't start their own sub-agents.
- Simultaneous tasks: each workspace runs up to 5 tasks at once, the same on every tier, and you can lower that in settings. Extra tasks wait in a queue (up to 20) and start automatically as slots free up.
- Long-running work: there's no time limit. Tasks, including long research, keep going until they're done, bounded by their credit budget instead of the clock. Work that needs to wait (a reply, a date, your answer) pauses and picks back up on its own.
- Daily limits: there's no cap on the number of tasks per day. On Agently's models, usage is limited by your credits. If you connect your own OpenAI or Anthropic key, runs on that provider use your key instead of credits, so you're limited only by your provider account, with a safety ceiling per task to protect against runaway spend. Routines can also have a daily spending cap. On the desktop app you'll be able to leverage your own Codex/Claude subscriptions as well.
- Rate limits: there's a general API rate limit and a per-workspace limit on actions in connected apps. If you use your own OpenAI or Anthropic key, that provider's rate limits also apply.
- What sets the limits: your credits depend on your tier. Concurrency, queueing and rate limits are the same for everyone, plus your model provider's limits when you use your own key.