Limits and quotas
This page lists the limits that apply across AlphaAgent Studio so you can plan around them. Limits fall into three groups: how many things each user can create, how large individual items can be, and runtime limits on chat and workflows.
Per-user resource counts
Each user can create up to 256 of each of the following:
| Resource | Limit per user |
|---|---|
| Agents | 256 |
| Swarms | 256 |
| Knowledge Graphs (AMPG) | 256 |
| Data connectors | 256 |
| Execution environments | 256 |
| Workflows | 256 |
Composition limits
| Limit | Value |
|---|---|
| Connectors attached to a single agent | 8 |
| Specialists in a swarm | 8 |
| Nodes in a workflow | 20 |
Field and content limits
| Item | Field | Limit |
|---|---|---|
| Agent | Name | 80 characters |
| Agent | Description | 1,000 characters |
| Agent | System prompt | 50,000 characters |
| Swarm | Name | 80 characters |
| Swarm | Description | 1,000 characters |
| Swarm | Routing instructions | 50,000 characters |
| Knowledge Graph | Name | 100 characters |
| Knowledge Graph | Description | 500 characters |
| Data connector | Name | 80 characters |
| Data connector | Description | 1,000 characters |
| Data connector | Usage guide | 50,000 characters |
| Workflow | Name | 80 characters |
| Workflow | Description | 1,000 characters |
| Execution environment | Name | 80 characters |
| Execution environment | Description | 500 characters |
| Conversation | Title | 200 characters |
Studio shows a live character counter as you approach these limits.
Document and file limits
Knowledge Graph documents
| Limit | Value |
|---|---|
| Maximum pages per PDF | 100 |
| Maximum size per document | 50 MiB |
Chat and workflow attachments
You can attach files in Express chat, Deep Traversal, and workflow document nodes alike — they are saved to the conversation store and analysed by the coder. Images are read with model vision.
| Limit | Value |
|---|---|
| Documents per request | 5 |
| Images per request | 20 |
| Maximum size per attached document | 4.5 MB |
| Maximum size per attached image | 3.75 MB |
| Allowed document types | .pdf, .doc, .docx, .xls, .xlsx, .ppt, .pptx, .txt, .md, .csv, .html |
| Allowed image types | .jpeg, .png, .gif, .webp |
Execution environment ranges
When you configure a custom execution environment, its resource settings must fall within these ranges:
| Setting | Range |
|---|---|
| Memory | 1,769–3,008 MB |
| Ephemeral storage | 512–3,008 MB |
| Timeout | 1–900 seconds |
Runtime limits
| Limit | Value |
|---|---|
| Maximum node executions in a single workflow run | 100 |
| Dataset retention (session datasets) | For the life of the conversation/run (no fixed expiry) |
Datasets that agents register during a conversation persist for the life of that conversation/run — there is no fixed expiry, so long, multi-step runs keep their datasets available throughout, and they remain available even after you leave and return to the conversation. The underlying data files in your workspace are governed by the conversation's workspace lifetime. See Datasets and data flow for how datasets and workspaces relate.
Timeouts
Timeouts work differently depending on what you're running:
| What's running | Time limit |
|---|---|
| A single code execution | Up to your execution environment's timeout — at most 15 minutes per run. See Execution environments. |
| Chat (Express and Deep Traversal) | No fixed turn timeout — the agent streams until it finishes the task. You can stop it yourself at any time. |
| Deep Traversal and multi-node Workflows | Designed for long, multi-step work and can run for extended periods (up to several hours). |
A single code cell is always capped at 15 minutes, but the overall chat or workflow keeps going across many cells and steps — so a long research run is not cut off by a fixed clock.
Deployment sizing and Bedrock quotas
These limits are set at deploy time and apply to the whole deployment rather than to a single user.
Fargate sizing tiers
Studio's services run on AWS Fargate, and you pick a sizing tier at install (and can change it on update). The tiers are xs, small, standard (the validated, recommended production size), large, and xlarge. Above standard the tiers add memory rather than CPU, because CPU is already at the Fargate 16-vCPU per-task ceiling.
ElastiCache node type
The ElastiCache (Redis/Valkey) node type that backs sessions and runtime cache is configurable at install and on update, with a sensible default. You can pick from a curated list or supply any valid node type.
Bedrock service quotas
Model inference runs on Amazon Bedrock. To avoid stalls during knowledge-graph ingest and Deep Traversal, raise your per-model Bedrock token quotas to the recommended floors for the models Studio uses — the runner model (Claude Sonnet 4.6), the planner model (Claude Opus), and the Cohere Embed model used for knowledge graphs. New AWS accounts often ship with low defaults, so this is checked during deployment preflight.
Workspace downloads
You can download a conversation's workspace as an archive. Very large workspaces may exceed the download size cap; in that case, narrow what you export (for example, download a single dataset or file rather than the whole workspace). See Code interpreter and datasets.
Notes
- These limits are enforced as you work, so you will usually see a clear message before an action is rejected.
- Limits apply per user; an organization's total capacity scales with its number of users.
- If you expect to need higher limits, contact your account team.
See also: Glossary and FAQ and troubleshooting.