Limits and quotas
Fixed in Studio
| Limit | Value |
|---|---|
| Built-in tools per agent | 5 |
| Workflows as a tool per agent | 1 |
| Documents in Fuentes (Sources) per agent | 5 |
| Tool calls per response | 1 to 5 (Herramientas (Tools) preset) |
| Maximum response length | 1 024 to 12 288 tokens |
| Message history | 0 to 6 messages in presets; customizable |
| Password | Minimum 8 characters |
| Items per page in the Agents API | Maximum 100 |
Per request (InferNode)
| Limit | Value |
|---|---|
Prompt plus max_tokens | Together they cannot exceed the model's context_window. The window is shared between the question and the answer, so max_tokens equal to the window always fails |
| Request duration | About 300 seconds; after that it responds 504 with provider_timeout |
| Usable response tokens | As many as the model manages to generate in that time. With a reasoning model, about 13 800 (Measured performance) |
| Window and output per model | GET /v1/models/{name}: context_window and max_output_tokens |
Per batch (InferNode)
Only if the installation has batches enabled. Guide in Process requests in batches. These values are the factory defaults, and the installation can change them.
| Limit | Default value |
|---|---|
| Input file size | 100 MiB; above that, 413 file_too_large |
| Line size | 1 MiB |
| Lines per file | 50 000 |
| Embedding inputs per file, summed across all lines | 50 000 |
custom_id length | 512 characters, not empty and unique within the file |
| Supported endpoints | /v1/chat/completions and /v1/embeddings, only one per file |
| Models per file | Only one |
| File storage per Client | 10 GiB by default; above that, 400 storage_quota_exceeded on upload |
| Active batches at once | Depends on the plan; above that, 429 too_many_active_batches |
| Queued lines | Depends on the plan; above that, the file fails with enqueued_request_limit_exceeded |
| Delivery time | No commitment. completion_window only accepts "24h" and is stored for compatibility |
| File retention | 30 days, for the input file and the output files. Each file carries its expires_at |
Depend on the Client's plan (InferNode)
| Limit | Where to see it |
|---|---|
| Requests per minute and per day | X-RateLimit-*-Requests headers; applied per organization, user and key |
| Tokens per minute and per day | X-RateLimit-*-Tokens |
| Images per minute | 429 with details.dimension: "ipm" |
| Spending caps (organization, user, key) | GET /v1/billing/balance → spending_caps |
| Credits | X-Credits-Remaining, GET /v1/billing/usage |
| Available models | GET /v1/models |
Depend on the installation (Labs)
| Limit | Where to see it |
|---|---|
| Sizes and default size | Nuevo laboratorio (New lab) form; GET /workbenches/v1/plan |
| Maximum per resource in custom mode | "Máximo:" (Maximum:) on each field; the plan's ceiling |
| Objects per lab (notebooks, volumes) | Rejection "tantos objetos de este tipo como permite su presupuesto" (as many objects of this type as its budget allows) |
| Disk classes | Only the default one |
| Client's total GPU capacity | Infrastructure dashboard |
Retention
| Data | Retention |
|---|---|
| Runs, messages and traces | Kept when you delete the agent or workflow; period per the agreement with the Client |
| Platform logs | 14 days by default |
| Access and recovery links | Single use, they expire |
| Idempotency key when creating agents | 24 hours |