@atomicstrata/llmwiki-local-workflows

The optional compiler-local workflow engine. This preserves llmwiki's existing workflow semantics: local execution per invocation, with run records and pending human gates retained between sessions. Persisted state does not imply automatic replay of interrupted actions. Applications needing automatic scheduling, worker recovery, distributed execution or managed approval routing supply a coordinator and call the standard llm-wiki-compiler SDK without starting local runs.

This publicly downloadable supporting package is not a separately supported consumer API. The following composition example is for compiler maintainers:

import { createLocalWorkflowHost } from "@atomicstrata/llmwiki-core/local-workflow-host";
import { createLocalWorkflowRuntime } from "@atomicstrata/llmwiki-local-workflows";
const runtime = createLocalWorkflowRuntime(createLocalWorkflowHost());

Install the exact matching @atomicstrata/llmwiki-core peer. The engine requires an explicit host and rejects a host from another core instance. Core retains persistence, locking and mutation authority; the engine requests effects through that host. Constructing the engine does not grant approval or trusted-write permission.

These low-level integration exports assume trusted in-process callers; they are not a sandbox for plugins. In particular, a caller can supply a human actor to a low-level gate operation. Use the standard interactive CLI for terminal-confirmed human approval; the standard SDK refuses programmatic human approval as before.

Applications should use llm-wiki-compiler, whose CLI and createWiki SDK compose both packages automatically, including applications supplying their own orchestration.

Build from the repository root with npm run build. Requires Node 24 or later.