humanlayer/12-factor-agents

What are the principles we can use to build LLM-powered software that is actually good enough to put in the hands of production customers?

TypeScript22531 ★AI Agent FrameworksGitHub ↗
Quality: weak 11/24
PAI: study 0.63
Appraised: 2026-05-27 current
Contents

Overview

Verdict

Rating Summary
Quality weak (11/24) Massive adoption and strong engineering signals are undercut by no releases, an 8-month commit gap, no license, and a README that navigates rather than teaches.
PAI Relevance study (0.63) Factor 12 (stateless reducer) and Factor 3 (own your context window) map directly onto PAI's Algorithm loop and context subsystem architecture — worth deep study, not installation.

Quality Assessment

11/24 — dormant-or-abandoned / under-documented / solid

Health: 2/8 (dormant-or-abandoned)

Failed:

Passed:

Documentation: 4/8 (under-documented)

Failed:

Passed:

Engineering Signals: 5/8 (solid)

Failed:

Passed:

PAI Relevance

Dimension Score Assessment
Harvest Value 2 Factor 12 (stateless reducer) maps directly onto PAI's Algorithm 7-phase execution loop design; Factor 3 (own your context window) is directly applicable to PAI's file-based memory and context-building patterns across WORK/, LEARNING/, and KNOWLEDGE/.
Integration Readiness 1 TypeScript aligns with PAI's stack but this repo is a documentation guide, not a bun-installable package; the planned create-12-factor-agent CLI scaffolding does not yet exist, so integration requires manually extracting and applying patterns rather than importing code.
Overlap Risk 1 Partial overlap with PAI's Loop skill (iterative agent cycles), Agents orchestration subsystem, and the Algorithm's observe-verify-learn structure — the principles mirror PAI's existing design without being redundant.
Gap Fill 1 PAI has no codified architectural principles document; the 12-factor vocabulary (stateless reducer, compact errors into context, unify execution and business state) would formalize implicit design decisions currently embedded across 46 hooks and 27 agents without a central reference.

Composite: 0.63

What Next

Landscape Position

Category: AI Agent Frameworks

In this category: VoltAgent--voltagent (excellent, integrate)

Standing: 12-factor-agents is the conceptual reference layer — a principles guide — while VoltAgent--voltagent is the runnable platform; they occupy complementary positions and neither obsoletes the other.

Evidence Base

Density: 6/10 — README (first 8KB) available; star/fork/issue counts available; creation and last-commit dates available; language confirmed; dependency manifest not available; CI configuration not available; individual factor content files not available; release history absent

Notes

The 8-month commit gap (last commit Sep 2025, appraisal May 2026) is notable given this repo's trajectory — either the guide is considered complete, or it has quietly stalled. The lack of any formal license (NOASSERTION) creates an ambiguity for any derivative work or create-12-factor-agent CLI built on top of it, despite the README badges citing Apache 2.0 for code and CC BY-SA 4.0 for content. The principles themselves — particularly Factors 3, 6, 7, 8, and 12 — are worth studying as a design vocabulary independent of whether the repo is actively maintained.