OpenExpertise vs Inngest / Temporal

OpenExpertise is a lightweight, LLM-aware SOP runtime; Inngest and Temporal are production-grade, general-purpose durable workflow engines that also happen to support LLM calls.

What Inngest and Temporal are

Temporal is a battle-hardened open-source durable workflow platform. Workflows are code (Go, Java, TypeScript, Python) and the Temporal server acts as a durable execution substrate: if a worker crashes mid-workflow, the workflow resumes from its last checkpoint. Temporal is widely used in large-scale backend systems — payments, logistics, ML pipelines — where reliability over days or weeks is non-negotiable. It has a rich ecosystem of SDKs, activity workers, and the Temporal Cloud managed service.

Inngest is a TypeScript-first event-driven durable functions platform. Functions are ordinary TypeScript decorated with inngest.createFunction, and the Inngest platform handles retries, concurrency limits, fan-out, scheduling, and event-driven triggers. It is designed to deploy alongside serverless (Vercel, Cloudflare Workers) and traditional backends alike. Inngest has strong LLM-aware primitives in its newer releases: step.ai.infer() provides structured retry and rate-limit handling for LLM calls, addressing the 429-problem that OpenExpertise also solves independently.

Both Inngest and Temporal are general-purpose engines. An LLM call is just another activity or step — they have no concept of "agent output schema," "state blackboard," or "evolution advisor." Their strength is operational reliability and scale; their abstraction layer is one level below LLM orchestration.

Where they overlap

• Durable execution with automatic retry on failure.
• Observable runs with event logs and step-level status.
• Parallel step execution.
• Resumability — a run can be paused and continued later.
• Structured state (inputs and outputs per step).

Where OpenExpertise differs

Dimension	OpenExpertise	Inngest / Temporal
LLM-aware primitives	agent node kind with structured output, prompt files, schema validation	LLM calls are plain activities/steps; no structured output or prompt abstractions
Deployment footprint	Runs as a CLI against a local SQLite store; zero separate services	Requires Inngest Dev Server / Temporal server or cloud service
Graph definition	Declarative YAML — the graph is a data artifact	TypeScript / Go / Java code — the workflow is a code artifact
Schema validation	state.schema validated with AJV before execution	Workflow inputs typed via language type system; no pre-run validation of the graph
Self-evolution	oe evolve proposes graph changes from runtime traces	Not present
Multi-CLI integration	cli-agent delegates to Claude Code / Codex / Gemini subprocesses	Not present
Authoring	oe ultra auto-generates YAML from natural language	Code-first
Scope	LLM + expert-process orchestration	General-purpose distributed systems: payments, ETL, ML, logistics
429-aware retry	Built in; exponential backoff on HTTP 429	Inngest has step.ai.infer(); Temporal requires custom activity retry policy

When to pick Inngest / Temporal over OpenExpertise

• You need production-grade durability at scale — tens of thousands of concurrent workflow instances.
• Your workflows run for days or weeks and must survive server restarts, deployments, and infrastructure failures.
• You're orchestrating general backend logic (payments, emails, ETL) that includes some LLM calls as steps.
• You need Temporal's rich observability UI, query language, and versioning for long-lived workflows.
• You need Inngest's event-driven triggers, serverless deployment, or deep integration with Vercel/Next.js.

When to pick OpenExpertise over Inngest / Temporal

• Your workflow is a team SOP — a fixed set of phases with LLM judgment at specific decision points.
• You want zero infrastructure: oe run examples/review-branch with a local SQLite store and no cloud account.
• You need the evolution loop: runtime trace → advisor → graph improvement proposal.
• You want a YAML artifact that non-engineers can review in a PR.
• You want to call Claude Code, Codex, or Gemini as first-class subprocess workers from within the graph.
• You want one-keyword authoring: oe ultra "triage incoming support tickets".

Specific positioning

Inngest and Temporal are load-bearing production infrastructure; OpenExpertise is a lightweight SOP runtime that trades raw operational scale for LLM-aware primitives and a self-improvement loop. If you need both, use Inngest/Temporal as the durable substrate and invoke OpenExpertise experiences as steps within that larger system.