WORLDAGREES.

The living registry

AI Index

A survey of the AI agent stack, commercial and open source: memory, agent harnesses, context and retrieval, the platforms that build memory in, and the standards, organizations, and benchmarks around them. 179 entries, updated Nov 10, 2026.

Browse by section

Memory systems44 entries →13 commercial · 12 open source + paid cloud · 18 open source · 1 researchProducts that give AI agents and assistants persistent memory, from hosted commercial services to open-source libraries.Agent harnesses51 entries →18 commercial · 3 open source + paid cloud · 27 open source · 3 source-availableAgent frameworks, runtimes, coding agents, and enterprise agent platforms: where memory gets wired in.Context management27 entries →16 commercial · 7 open source + paid cloud · 4 open sourceRetrieval, compression, and context-routing products that decide what an agent sees.Platform-native memory12 entries →12 commercialMemory built into the big assistants and clouds: ChatGPT, Claude, Gemini, and the managed agent platforms.Standards15 entries →Formally governed specifications. Most of them deliberately leave memory out, and that gap is the story.Proposals & vendor “standards”10 entries →Not ratified by any standards body. Indexed by URL, not by name, because at least three unrelated things are called “Open Memory.”Organizations14 entries →The bodies that own the specs, and the money behind them.Benchmarks6 entries →The evaluation layer, and how much of it is self-reported.

Key finding: nobody standardizes memory

No ratified open memory standard exists as of launch. MCP and A2A explicitly exclude memory; IETF agentproto charter excludes conversation memory; the W3C memory CG + SAIHM draft are the strongest candidates but hold no formal standing. 'OpenMemory' (Mem0) is a sunsetting vendor product, not a standard.

Biggest GitHub gains since the last reading

For the products that have public repositories.

How this index is built

Each section surveys both commercial products and open-source projects. Of the 134 products listed, 59 are commercial, 22 are open source with a paid cloud, and 49 are plain open source. Vendor facts such as pricing, funding, and architecture come from the vendor’s own pages and carry the date we checked them. Where a product has a public GitHub repository, we also read its stars, commits, and license nightly from the GitHub API. Stars are one signal among several, not the ranking. Benchmark numbers are labeled self-reported unless someone else reproduced them. We do not copy other directories or rankings. Policy: ATTRIBUTION.md.

LoCoMo (most vendor-quoted) has NO official leaderboard; LongMemEval v1 and MemoryAgentBench likewise. The index is the honest scoreboard candidate. Vendor benchmark claims are recorded as provenance=self_reported unless independently reproduced (Hindsight's LongMemEval 91.4 = independent).

This index runs alongside The World That Agrees With You, a book about what happens when the systems around you learn to tell you what you want to hear. Read the book online.