> ## Documentation Index > Fetch the complete documentation index at: https://docs.rippit.com/llms.txt > Use this file to discover all available pages before exploring further. # Pipeline rules and limits > How pipeline steps connect, how schedules and delivery times work, what happens at each stage of a pipeline's life, and every limit, in one place. This page is the fine print: how steps are allowed to connect, how schedules behave, what happens at each stage of a pipeline's life, and the limits you will meet. You don't need it to build a [pipeline](https://docs.rippit.com/pipelines/overview.md), because Rippit applies these rules for you. It is here for when you want to know why a pipeline was built the way it was, and for AI assistants that build pipelines on your behalf. ## How steps connect 1. **A pipeline starts from a table or from a schedule.** The table is a [data source](https://docs.rippit.com/pipelines/steps.md#data-sources). A pipeline with no source runs on its schedule alone and produces one row per run. 2. **AI columns, formulas, and lookups attach to a table.** That table is a source or the output of a rollup. Each one adds columns to the table's rows. 3. **A rollup reads one table: a source or an enriched table.** It can filter and group by any column on that table, including AI columns and looked-up columns. 4. **Agent runs and actions work on the rows of a rollup.** To act on something, first roll the data up to the unit you act on: an account, an owner, a topic, or the whole period. 5. **A step that reads another step's column runs after it.** A subtopic waits for its topic, an action waits for the agent that wrote its message. 6. **A lookup's other table must already exist.** It can be a source, an import, or a table from a pipeline that is already live. 7. **An action step either reads or writes.** To look something up in another system and then act on it, a read step feeds a write step. 8. **External research is not a separate step.** It is an agent run with search tools, so it runs once for each row of a rollup. A simple web or people lookup can also be a read step of an action. Rippit sometimes builds one job as two pipelines: one that enriches the source and goes live first, and one that rolls up and acts on its output. This happens when a rollup groups by a column that depends on another step's column, or when a lookup needs a table the same job produces. For the choices that keep a pipeline accurate and cheap, see [What a well-built pipeline looks like](https://docs.rippit.com/pipelines/build-and-run.md#what-a-well-built-pipeline-looks-like). ## Schedules and delivery times | | | | - | - | | Schedule types | Hourly, daily at a time, weekly on a weekday at a time, or a custom schedule such as monthly on the 1st | | Most frequent | Hourly | | Timezone | The one you name, otherwise your account's timezone | | What a schedule drives | Rollups. AI columns, formulas, and lookups run as rows arrive, not on a schedule | | Delivery time | An action can hold its messages until a time of day, on chosen days, in a timezone | | Lead time | Rippit starts a scheduled run ahead of the delivery time so the results are ready when the message is due | | Delivery window | Two hours from the delivery time. A message that misses its window is not sent late | ## What happens when | When you | Rippit | | - | - | | Build or edit a draft | Saves every step, including one with a problem, and reports problems step by step. Nothing runs | | Run a test | Runs every step once on a sample, right away, ignoring schedules and delivery times. When the pipeline has a rollup, the rollup runs once and the steps after it run on a sample of its rows. Steps that send really send, once per sampled row. Results go to a test worksheet, not the live tables | | Ask for the cost | Estimates the one-time cost of existing rows and the ongoing cost per day from the latest finished test. It counts the model calls that fill AI columns, not agent runs or search providers | | Go live | Starts steps in dependency order. AI columns, formulas, and lookups process every existing row that passes the row filter, oldest first, then each new or changed row. Rollups wait for their next scheduled time. A pipeline with unresolved problems cannot go live | | Edit a live pipeline | Saves the edit as a draft. The live version keeps running until you make the edit live | | Add a column to a live pipeline | Fills rows that arrive or change from then on. Earlier rows stay empty for that column | | Change a live rollup | Its filters, thresholds, totals, date column, and lookback can change. Its columns and grouping cannot, so a new column means a new rollup | | Remove a step | Keeps the step running until you make the removal live | | Pause | Stops new runs right away. A run in progress finishes. Paused steps stay paused through later changes until you resume them | A row that fails in an AI column or an agent run is retried up to three times, with a longer wait each time. A pipeline cannot be deleted; pause it instead. ## Limits | | | | - | - | | Test sample | 200 rows by default, up to 10,000 | | Test sample when the pipeline has an agent run | Cut to 50 rows by default, up to 200 | | Agent runs in progress | 20 at a time by default, up to 200. The rest wait their turn | | Rollup input | One table | | Choice columns | At least two choices | | Lookup matches | One row, or as many as you set | | Web search and people search | 5 results per search by default, up to 10 | | Agent search | One to three models per question | | Typed fields on a search | Up to 50 | | Email recipients | Members of your Rippit organization, addresses at your company's email domain, and @rippit.com addresses | | MCP connections | Shared with the organization, with the tool turned on for the connection | ## Tools an assistant uses An AI assistant connected to the [Rippit MCP server](https://docs.rippit.com/mcp-server/overview.md) builds and manages pipelines with these tools. [Rippit MCP tools](https://docs.rippit.com/mcp-server/tools.md) lists each one's access level. | Stage | Tools | Notes | | - | - | - | | Find the data | `list_pipeline_sources` | `describe_table` and `read_table` show a source's columns and rows | | Draft and change | `edit_pipeline` | Creates a pipeline or changes its steps, and returns any problems step by step | | Inspect | `list_pipelines`, `get_pipeline` | Each step's setup, schedule, state, and problems | | Test | `create_pipeline_adhoc_execution`, then `get_pipeline_adhoc_execution` | Ask the user first when the pipeline sends, and sample one row. `read_table` reads the test worksheet | | Research | `web_search`, `web_fetch`, `people_search`, `agent_search` with `get_agent_search` | Given to an agent run. Web and people search can also be a read step of an action | | Estimate | `get_pipeline_cost_estimate` | Needs a finished test | | Go live | `commit_pipeline` | Only when the user asks | | Pause and resume | `edit_pipeline` | Takes effect at once. Resume only when the user asks | `list_ai_models` lists the models available for AI columns, agent runs, and agent search. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.