Piped Input

Compose praisonai run/code/chat in Unix pipelines and CI

Pipe any text into praisonai run, code, or chat and it is merged with your prompt before the agent runs.

Quick Start

1. Pipe a file into run

   cat error.log | praisonai run "Diagnose the root cause"
   
2. Pipe into code

   cat data.json | praisonai code "Write a parser for this shape"
   
3. Pipe into chat

   echo "$STACKTRACE" | praisonai chat "Explain this"
   

---

How It Works

The prompt argument and piped stdin are merged before dispatch.

Component	Role
resolve_cli_input(prompt)	Merges the prompt argument with piped stdin (prompt first).
read_stdin_if_available()	Non-blocking read with a 10 MB cap.
select.select guard	Prevents stalls when stdin is an open pipe with no EOF (common in CI).

---

Merge Order

The prompt argument comes first, then the piped body, joined with \n.

Prompt arg	Piped stdin	Final input to agent
Set	Set	prompt + "\n" + piped
Set	Empty / no pipe	prompt (unchanged)
None	Set	piped (becomes the prompt)
None	Empty / no pipe	None (interactive REPL if TTY)

---

Decision Diagram

praisonai run chooses one path per invocation — only the default text path merges piped stdin.

---

Skip Rules for praisonai run

praisonai run skips the stdin merge when merging a piped body would be meaningless.

• YAML target — an existing .yaml / .yml file (case-insensitive, so AGENTS.YAML counts). Concatenating a piped body into a YAML path would be nonsense.
• --agent — a named custom agent runs its own definition.
• --command — a named custom command uses TARGET as $ARGUMENTS.
• --restore — the command exits before ingestion, so ... | praisonai run --restore last never drains the pipe.

praisonai code and praisonai chat call resolve_cli_input(prompt) unconditionally at the top of the handler, so they always merge piped stdin.

---

Size Cap & Platform Notes

Piped input is capped at 10 MB. Larger streams are truncated to 10 MB before merging.

Windows piped stdin works as of PraisonAI 2026-07-07 (PR #2705). Because Python’s select.select is socket-only on Windows, the reader now uses a stat-based pipe classifier instead. Both cmd and PowerShell redirection are supported:

type error.log | praisonai run "Diagnose the root cause"

Get-Content error.log | praisonai run "Diagnose the root cause"

Interactive terminals with no redirection continue to skip the stdin read, so a bare praisonai run still drops into the REPL. On Unix the existing select.select() guard is unchanged. To attach a file explicitly instead, --file still works on praisonai code / praisonai chat:

praisonai code --file error.log "Diagnose the root cause"

---

CI / Scripting Examples

GitHub Actions

- name: AI review of the diff
  run: git diff origin/main...HEAD | praisonai run "Review these changes for bugs"
  env:
    OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}

kubectl logs deploy/api --tail=200 | praisonai run "Explain these errors"

curl -s https://api.example.com/log | praisonai code "Add tests for the failing calls"

---

Best Practices

Put the prompt first, pipe the context second

The merge order is prompt then piped body, so a leading prompt produces the clearest instruction for the agent.

cat error.log | praisonai run "Diagnose the root cause"

For YAML agents, do not pipe — use --file or an inline instruction

The YAML skip rule means a pipe into a YAML target is silently ignored. Pass context with --file or write the instruction inline.

praisonai code --file config.yaml "Validate this config"

In CI, do not pipe more than 10 MB — split large logs

The 10 MB cap truncates silently. Trim or tail large streams before piping.

tail -n 500 huge.log | praisonai run "Summarize the recent failures"

Prefer named custom agents for repeatable pipelines

A named agent (--agent) keeps repeatable pipelines cleaner than reshaping stdin every invocation.

praisonai run --agent reviewer "Audit the latest changes"

---

Related

Run

Run agents from files or prompts.

Code

Code assistant mode for programming tasks.

Chat

Interactive chat mode with AI agents.

CLI Dispatch

Command dispatch overview.