100+ AI Agent & RAG apps you can actually run — clone, customize, ship.
INTEGRATE — Add to vault as the canonical runnable-templates reference for LLM agent and RAG pattern work; clone individual templates to bootstrap new PAI components.
Formula verdict: STUDY (pai_harvest = 2 AND standalone = 15 < 16). Recommended verdict: INTEGRATE. Reason: standalone = 15 is one point below the INTEGRATE threshold of 16; all other INTEGRATE criteria are met (pai_fit = 0.75 >= 0.5, pai_overlap = 0 <= 1). The one-point shortfall is a structural artifact of a monorepo template collection — no top-level release tag and no root-level manifest are expected and not quality deficits. 110,043 stars over ~25 months, 4-day-old last commit, and 0.75 PAI fit justify the upgrade.
15/24 — maintained / adequately-documented / early-or-minimal
Failed: H1: FAIL — latest_release is "none"; no tagged releases exist H2: FAIL — no releases at all, so no recent-release check possible H8: FAIL — README contains star/fork/license badges but no CI badge (GitHub Actions, Travis, CircleCI) and no .github/workflows reference visible
Passed: H3: PASS — last commit 2026-05-09, 4 days before appraisal date H4: PASS — last commit 4 days ago, well within 30-day window H5: PASS — archived: false H6: PASS — 3 open issues, satisfies >0 AND <100 healthy-triage range H7: PASS — Apache-2.0 license present
Failed: D5: FAIL — no heading containing "API", "Configuration", "Options", "Reference", "Commands", or "Parameters" visible in 8KB README excerpt D8: FAIL — no "Limitations", "Caveats", "Known Issues", "Trade-offs", or "Not supported" section present
Passed: D1: PASS — README content is present and extensive D2: PASS — README is many KB with 13-category table of contents, detailed project listings, and prose sections D3: PASS — Quick Start section contains explicit "pip install -r requirements.txt" command D4: PASS — Quick Start code block with git clone, pip install, and streamlit run under a named section D6: PASS — first 500 characters include "100+ AI Agent & RAG apps you can actually run — clone, customize, ship" clearly stating purpose D7: PASS — multiple links to https://www.theunwindai.com for step-by-step tutorials
Failed: E1: FAIL — primary language is Python, not in typed-language list (TypeScript, Rust, Go, Java, Kotlin, C#, Swift, Scala, Haskell) E2: FAIL — dependency manifest field reports "Not available"; per-project requirements.txt files exist in subdirectories but no root-level manifest E3: FAIL — cannot assess dependency count without a top-level manifest E4: FAIL — no mention of test framework, pytest, CI test step, or test script in README excerpt
Passed: E5: PASS — 110,043 stars far exceeds the 50-star threshold E6: PASS — created 2024-04-29, ~25 months to 2026-05-13; 110,043 / 25 ≈ 4,402 stars/month, far above 2/month floor E7: PASS — 16,297 forks far exceeds the 5-fork threshold E8: PASS — description is 64 characters, clearly describes content and value proposition
| Dimension | Score | Assessment |
|---|---|---|
| Harvest Value | 2 | The collection spans MCP agents, voice AI agents, multi-agent teams, RAG pipelines, memory-enabled apps, and fine-tuning — each category directly relevant to PAI design concerns. Provider-agnostic scaffold patterns (Claude/Gemini/GPT/Llama swap via config) and self-improving agent skills are directly harvestable architectural patterns. |
| Integration Readiness | 1 | Individual templates are self-contained and clone-ready in 3 commands per README, but the repo is a template collection not a library — adapting any specific template to PAI context requires hours of glue work and no root-level import/install path exists. |
| Overlap Risk | 0 | First-in-category in the vault; no existing vault repo covers runnable LLM app templates. Conceptual thematic overlap with karpathy--autoresearch (agent loops), VoltAgent--voltagent (agent engineering), and humanlayer--12-factor-agents (agent principles) but none provide runnable code templates. |
| Gap Fill | 1 | The landscape shows thin coverage across Multi-Agent Orchestration (2 repos) and AI Agent Engineering (1 repo) with no runnable reference implementations; this fills the practical "how do I actually build this" gap that principle and framework repos leave open, though the Gaps section does not name this category explicitly. |
Composite: 0.75
Category: LLM App Templates Crowding: 0 repos in vault (first-in-category) Alternatives: first in this category vs. top alternative: No direct vault alternative exists; karpathy--autoresearch is the closest adjacency but covers only autonomous ML research loops, not a general runnable-template cookbook. Landscape impact: filling a gap — no existing vault repo provides original, end-to-end-tested, runnable source templates across the full modern LLM stack
Density: 8/10 — Available: README content, stars and forks, license, creation and last-commit dates, topics, primary language, archive status, open issue count, release tag status. Missing: dependency manifest (no root-level file provided) and CI configuration details (no workflow file content available).
The low engineering score (4/8) is an artifact of the monorepo template-collection structure: Python is not in the typed-language list, there is no root-level manifest (each of 100+ subdirectories ships its own requirements.txt), dependency count is unmeasurable at the root, and there is no centralized test suite. These are expected structural choices for a cookbook, not quality deficits. Social proof (110k stars, 16k forks, Trendshift featured) and recency (4-day-old last commit, active additions through May 2026) strongly indicate ongoing health beyond what the health probes capture. When cloning individual templates, confirm each subdirectory requirements.txt is current before use.