TL;DR
The limits on what one Asset ↔ Term Linker run and one agent can read and write, and what happens when a limit is reached.
Your AI can read this via Docs MCPcurl -fsSL "https://docs.atlan.com/install-docs-mcp" | bashConnect
Asset ↔ Term Linker limits Private Preview
Limits that apply to Asset ↔ Term Linker during private preview. They're raised over time.
Scope limits
A run that exceeds a scope limit is refused before it starts, and the message names the limit.
| Limit | Value | When exceeded |
|---|---|---|
| Glossaries per agent | 10 | The agent can't be saved |
| Asset collections per agent | 1 | Only one can be selected |
| Verified terms per run | 500 | The run is refused |
| Tables, views, and materialized views per run | 100 | The run is refused |
| Tables, views, and columns per run | 3,000 | The run is refused |
Write limits
| Limit | Value | When exceeded |
|---|---|---|
| New links per run | 250 | The highest-confidence links are written, and the rest are listed as over the limit. A later run links the rest |
| Links per agent in 24 hours | 1,000 | Publishing is refused until the window clears |
| Links per term | 10 | The excess, lowest confidence first, moves to review |
| Age of a sample you can publish | 7 days | Publishing is refused |
Decision limits
| Limit | Value | Effect |
|---|---|---|
| Candidate assets per term | 15 | Lower-ranked candidates aren't considered |
| Run instructions | 1000 characters | Enforced in the setup flow |
Run limits
| Limit | Value |
|---|---|
| Scheduled runs | Not supported |
| Concurrent runs per agent | 1 |
See also
- Link terms to assets: plan the scope, create the agent, sample, review, and publish
- Understand Asset ↔ Term Linker: how links are decided