Build AI methods with your coding agent, then run them anywhere: as an MCP for chatbots, as a webapp for people, or via API for your software.
# Pipelex-API deployment config — the packaged default, loaded by `load_api_config()`
# (pipelex_api/api_config.py) via core's env-aware `load_plugin_config` and validated into `ApiConfig`.
# Keys live at the file root (no `[api]` wrapper) — `load_plugin_config` validates the whole merged
# document against the schema, exactly like `pipelex-temporal`'s `temporal.toml`. The packaged
# default here is deep-merged with an optional `api_{env}.toml` / `api_override.toml` at `~/.pipelex`
# then the project `.pipelex` (env selected by `PIPELEX_ENV`). This source-available base names NO
# orchestrator: it ships the in-process `direct` mode and refuses per-request override. A deployment
# flavor bakes its own `.pipelex/api_{env}.toml` to flip this (e.g. `pipelex-api-hosted` sets
# `orchestration_mode = "temporal"`).
# Which orchestrator a top-level run dispatches to, through the orchestrator registry.
# `orchestration_mode` is an OPEN string token: core owns "direct" (in-process); each orchestrator
# plugin owns its own ("temporal" from pipelex-temporal, "mistral-workflows" from
# pipelex-mistral-workflows). A token whose plugin is not installed fails loud at dispatch with the
# plugin's install hint. The delivery axis (blocking vs fire-and-forget) is NOT configured here — it
# is set by the endpoint (`/execute` and `/validate` block; `/start` is fire-and-forget).
orchestration_mode = "direct"
# Whether a caller may override `orchestration_mode` per request. Off on the base (and recommended off
# on hosted flavors): a locked-down distributed runner must not be coercible into `direct`.
allow_request_orchestration_mode_override = false