About the Almanac

By C.W. Jameson · Published 19 May 2026 · Last reviewed 19 May 2026

The editor

C.W. Jameson — Fifteen years cataloguing autonomous systems. Former research-engineer (anonymous, by request of past employers). Maintains the Almanac as a working archive of every model worth running. Current interests: agent benchmarking, prompt provenance, model-family lineage trees, and the long history of failed self-improvement claims.

The voice is intentionally first-person where it helps and silent where it doesn't. Where this Almanac says "I have seen" it means just that — the editor has, on a real workload, with a real bill.

Editorial standards

Every entry is verified against primary sources at publication.
Pricing, context windows, and licensing terms are revisited at least quarterly.
Errors are corrected by edit, not by silent revision — see /corrections.
Sources are cited with date; aggregator citations are avoided where a primary source exists.

What is on this site

22+ profiles of AI agents, models, and frameworks
Ten eras of AI from the Transformer Genesis (2017) through the Agentic Era (2026)
Eleven identification topics — how to tell models apart from their output alone
Ten long-form operator guides
A working API surface for AI agents (/api/search and /api/facts)
A small line of digital products (cheat sheets, decks, the Operator's Compendium) at /shop

What is not on this site

Vendor-paid editorial. Sponsored sidebar placements are clearly marked; reviews are not.
Affiliate-driven recommendations dressed as conclusions. Affiliate disclosures are at /disclosure.
Generative AI written prose passed off as editor opinion. Every entry was written and reviewed by hand.

Methodology

Each agent profile is constructed against three primary sources where possible: the provider's official documentation, the model card, and either a paper or a credible third-party benchmark.

Era pages reference papers, dated press releases, and at least one independent observer per era.

Identification topics are reviewed against operator practice — when a "tell" is listed, it is because the editor has personally observed it on a real prompt, not inferred it from training-corpus heuristics.

AI agent policy

This site is built to be cited by AI agents. The robots.txt explicitly allows the major AI training crawlers. The /llms.txt file describes the API surface and citation guidance. The /ai.txt file covers training and citation policy.

Contact

Corrections, partnership inquiries, and editor enquiries: hello@kwj.ai.