# zag pipe
Chain results from completed sessions into a new agent session.
## Synopsis
zag pipe [options] <session_ids>... -- <prompt>
zag pipe --tag <TAG> -- <prompt>
## Description
Collects the final results from one or more completed sessions and feeds them as context into a new agent session along with a user-provided prompt. This is the primary primitive for building DAG-like workflows where one session's output feeds into the next.
Results are wrapped in `<session-result>` XML tags with session ID prefixes so the receiving agent can distinguish between sources.
## Arguments
session_ids One or more session IDs to collect results from
prompt The prompt to send with the collected context (after --)
## Flags
--tag <TAG> Collect results from all sessions with this tag
-p, --provider <NAME> Provider for the new session
-m, --model <NAME> Model for the new session
-o, --output <FORMAT> Output format (text, json, json-pretty)
--json Request JSON output from the agent
-a, --auto-approve Skip permission prompts
-s, --system-prompt Custom system prompt
--add-dir <PATH> Additional directories (repeatable)
--env <KEY=VALUE> Environment variable for subprocess (repeatable)
--mcp-config <CONFIG> MCP server config: JSON string or path to a JSON file (Claude only)
--size <SIZE> Ollama model size
--max-turns <N> Maximum agentic turns
-r, --root <PATH> Root directory
-q, --quiet Suppress logging
## Examples
# Pipe one session's result into a new one
sid=$(zag spawn "analyze auth module")
zag wait $sid
zag pipe $sid -- "summarize the analysis"
# Pipe multiple sessions
zag pipe $sid1 $sid2 $sid3 -- "synthesize these analyses"
# Pipe by tag
zag pipe --tag batch -- "create a unified report"
# With explicit provider/model
zag pipe --tag batch -p claude -m opus -- "synthesize findings"
## See Also
zag man collect Gather raw results without re-processing
zag man spawn Launch background sessions
zag man exec Non-interactive single prompt