Plan limits
An empty cell means the limit is unset — there is no cap. A zero in a numeric limit reads the same way: “no limit”, not “nothing allowed”.
| Limit | Free | Start | Pro | Business |
|---|---|---|---|---|
| Bring-your-own LLM connections | — | ✓ | ✓ | ✓ |
| Public chat | ✓ | ✓ | ✓ | ✓ |
| White-label branding | — | — | ✓ | ✓ |
| Allowed connection kinds | unset (no limit) | unset (no limit) | unset (no limit) | unset (no limit) |
| Deploy channels | web, telegram, vk, webhook, cron | web, telegram, vk, webhook, cron | web, telegram, vk, webhook, cron | web, telegram, vk, webhook, cron |
| Allowed models | unset (no limit) | unset (no limit) | unset (no limit) | unset (no limit) |
| API requests per minute | 30 | 60 | 300 | 1,200 |
| Renewal period | every 30 days | every 30 days | every 30 days | every 30 days |
| Quarks per period | 300 | 7,000 | 20,000 | 100,000 |
| Minimum cron interval, s | 1,800 | 600 | 300 | 60 |
| Collection storage | 100 MB | 1 GB | 10 GB | 100 GB |
| Run history retention, days | 3 | 14 | 30 | 90 |
| File storage | 100 MB | 1 GB | 10 GB | 100 GB |
| Knowledge-base storage | 100 MB | 1 GB | 10 GB | 100 GB |
| Document collections | 5 | 20 | 100 | 1,000 |
| Concurrent nodes per run | 3 | 6 | 10 | 20 |
| Concurrent runs | 1 | 5 | 20 | 50 |
| Connections | 3 | 20 | 100 | 500 |
| Published scenarios | 1 | 10 | 50 | 200 |
| Files | 100 | 1,000 | 10,000 | 100,000 |
| Knowledge bases | 2 | 10 | 50 | 500 |
| Loop iterations | 100 | 500 | 1,000 | 5,000 |
| Members per workspace | 1 | 3 | 10 | 100 |
| Model price ceiling, quarks per 1000 output tokens | 3 | unset (no limit) | unset (no limit) | unset (no limit) |
| Node executions per run | 200 | 1,000 | 2,000 | 5,000 |
| Subgraph recursion depth | 3 | 5 | 10 | 15 |
| Chat session queue depth | 3 | 5 | 10 | 20 |
| Single upload size | 10 MB | 25 MB | 25 MB | 25 MB |
| Vector points | 50,000 | 1,000,000 | 5,000,000 | 50,000,000 |
| Run time ceiling, s | 120 | 300 | 900 | 1,800 |
| Saved versions per scenario | 5 | 20 | 100 | 500 |
| Scenarios | 5 | 30 | 200 | 1,000 |
| Workspaces | 1 | 3 | 10 | unset (no limit) |
| Messages per chat | 500 | 1,000 | 2,000 | 5,000 |
| Messenger history retention, days | 7 | 14 | 30 | 90 |
| Node access | all but a few | all but a few | all but a few | all but a few |
| Allow negative quark balance | — | — | ✓ | ✓ |
How the limits are applied
- Quarks and concurrency are checked when a run is admitted: too low a balance and the run is rejected; a full concurrency quota and it waits in the queue.
- Storage quotas are counted per owner, not per workspace: extra workspaces do not add extra room.
- Deploy channels are checked at deploy time and are open on every plan. The difference is cadence: the lower the plan, the longer the minimum interval between schedule fires — the exact values are in the table above.
- Retention decides how long runs and logs live; after that a background sweep removes them and the UI marks the run as purged.
What next
- Quarks and cost — what a run costs.
- Usage — where to see spend and quotas.