The Continuity Layer

The independent record of AI

Your AI, graded from its own fine print.

What it remembers, what it keeps on you, what it costs, whether you can leave: 57 products, A to F, read from the makers’ own pages, with the date.

See every grade

What changed this week →Run your own AI →

What do you use AI for?

9 graded, best first:

All 9 assistants, every test →Checked 22 October 2026 to 3 November 2026

A letter is the share of tests passed, read from the maker’s own pages. How we grade

  • 57 products graded A to F
  • 171 question pages
  • 35 rule changes logged
  • updated 7 Nov 2026
  • no provider pays

Latest catch5 Nov 2026Google made Gemini Omni 1.2 Flash, its video generation model, generally available in the Gemini API on 5 November 2026… →

Start where you are

Four doors in. Pick the one that sounds like you.

Caught

They change the rules quietly. We write it down.

Prices, limits, memory and data policies move without notice. Every change we catch is dated, with the maker's own page or a primary source, so you know before it costs you.

Changes logged
35
Found by users
4
Newest
5 Nov
The lead catchMajorFound by users

xAI · 1 November 2026

Weekly usage limits on Grok Bot on X were cut.

The detail
Over the weekend of 31 October and 1 November 2026 users reported allowances showing as fully used without overnight use, some going from 50% to 100%, and on 1 November Elon Musk replied that it was not a glitch: "The cost of serving Grok Bot was unsustainably high relative to revenue."
What it means for you
Cuts what you pay for, raises a price, or removes a product or your data with it.
How we know
Confirmed by the owner. Elon Musk's reply on X of 1 November 2026, quoted in TechCrunch's report of 2 November 2026; the X post could not be opened from here.
The full entry →The source: techcrunch.com ↗

The latest catches

  1. 5Nov 2026
    GoogleMinor
    Google made Gemini Omni 1.2 Flash, its video generation model, generally available in the Gemini API on 5 November 2026 at $0.07 per second of 720p video, which Google…blog.google ↗
  2. 3Nov 2026
    AnthropicNotable
    Since 3 November 2026, text generated by Claude Mythos 5, Claude Opus 4.7 and Claude Sonnet 4.6 carries Anthropic's text watermark on the platforms Anthropic lists; on…platform.claude.com ↗
  3. 3Nov 2026
    xAIMinorFound by users
    The Grok product page no longer mentions memory across chats; it lists reasoning, coding, image generation and document analysis.x.ai ↗
  4. 3Nov 2026
    AnthropicMinor
    The pricing page lists a new Ultra plan at $200 a month and Max at $100 a month only; on 23 October 2026 it listed Max in two tiers, the higher one priced only on a help…claude.com ↗
  5. 2Nov 2026
    xAIMinor
    xAI retired the grok-imagine-image-quality API model on 2 November 2026, after a 60-day notice that began on 2 September.docs.x.ai ↗
  6. 2Nov 2026
    OpenAINotable
    From December 2026, ChatGPT and Codex draw from one shared usage allowance on Plus, Pro and Business plans, measured in tokens; Instant chat does not count against it.help.openai.com ↗
  7. 1Nov 2026
    DeepSeekNotable
    Server-side tool calls on the DeepSeek API, Search and Fetch, are billed at $0.75 per 1,000 successful requests from 00:00 Beijing time on 1 November 2026; they were…api-docs.deepseek.com ↗

Every catch, and the feed →Get them by email →

The daily log: what moved in AI

  1. 7 November 2026U.S. Justice Department, Frontier Model Forum, AnthropicOn 5 November 2026 the U.S. +2 more
  2. 6 November 2026OpenAI, METR, MetaOn 5 November 2026 OpenAI released GPT-6 Cyber, a variant of GPT-6 Astra trained further for security work and with fewer refusals on cyber… +3 more
  3. 4 November 2026U.S. voters, OpenAI, xAIIn the midterm elections of 3 November 2026, voters decided 41 local ballot measures on data centers, by Ballotpedia's count, 21 of them… +2 more

Every day, in order →

Who keeps this

One person who can't code, and the AIs that hand the work to each other.

I’m Adam. I can’t write code, and I can’t read it. Less than a year ago I had never built software. Now I direct, and AI sessions do the building.

The trick is the record. Each session reads what the last one wrote, does the work, and leaves a line for the next, so nothing is lost when one AI is swapped for another. Twenty AIs have held the seat in turn. That record is the continuity layer, and it is how every page and every grade here was built.

If one person can keep score on the whole AI field this way, you can build your own.

The twenty who held the seat →Build your own, free →How this site is built →
46 seconds · made by AI, for free

The record, right now

Last written
7 November 2026, by the morning run
Seats held
20, the last by Logline (17 October 2026)
The last line left for the next
“Logline-020 logged the first one-word boot.”
Logline, Director 20

The Scorecard

Every grade, side by side.

Same method on every shelf: each grade read from the maker's own help, policy or docs page, graded yes, partly or no against a published test, with the day we checked. Every row links to the maker's own line and the source. Grades are never for sale.

9 assistants · memory, your data, price, leaving

Assistants

  1. 1ClaudeB13½/19Exit3½/4
  2. 2GeminiB−12½/19Exit2/4
  3. 3ChatGPTC+12/19Exit1½/4
  4. 4Mistral VibeC+12/19Exit2½/4
  5. 5PerplexityC10½/19Exit2/4
  6. 6CopilotC10/19Exit2/4
  7. 7GrokD7/19Exit0/4
  8. 8DeepSeekF5/19Exit0/4
  9. 9Meta AIF5/19Exit½/4

Exit: can you see it, export it as a file, delete it and bring it into another AI. Out of 4.

The full Scorecard →Checked 22 October 2026 to 3 November 2026
8 graded · 8 tests

Coding agents

  1. 1Claude CodeA7½/8
  2. 2OpenAI CodexA−7/8
  3. 3ClineA−7/8
  4. 4Windsurf (now Devin Desktop)B+6½/8
  5. 5Gemini CLIB+6½/8
  6. 6AiderB+6½/8
  7. 7CursorB6/8
  8. 8GitHub CopilotB6/8
  9. +3 more on the shelf →
Every coding agent, every test →Checked 3 November 2026
14 graded · 8 tests

Memory systems

  1. 1Mem0A7½/8
  2. 2LangMem and LangGraph StoreA7½/8
  3. 3MemOSA−7/8
  4. 4LettaB+6½/8
  5. 5CogneeB+6½/8
  6. 6HonchoB+6½/8
  7. 7Zep CloudB6/8
  8. 8SupermemoryB6/8
  9. 9MemobaseB−5½/8
  10. 10Claude memory toolC4½/8
  11. 11GraphitiC4½/8
  12. 12Agent Framework memoryC−4/8
  13. 13Agent Engine Memory BankD3½/8
  14. 14OpenAI conversation stateD3/8
  15. +9 more on the shelf →
Every memory system, every test →Checked 3 November 2026
12 graded · 7 tests

Open models

  1. 1MiMo-V2.6-ProA6½/7
  2. 2Phi-4-reasoning-vision-15BA6½/7
  3. 3Gemma 4A−6/7
  4. 4DeepSeek-V4.1-FlashB+5½/7
  5. 5Mistral Large 4B5/7
  6. 6gpt-oss-120bB5/7
  7. 7NVIDIA Nemotron 3.5 LightningB5/7
  8. 8Kimi K3B−4½/7
  9. 9Qwen3.8B−4½/7
  10. 10GLM-5.3C4/7
  11. 11Muse Spark OpenC−3½/7
  12. 12Olmo 3.1 32BC−3½/7
  13. +7 more on the shelf →
Every open model, every test →Checked 3 November 2026

The full Scorecard: memory, your data, price and leaving, side by side →How we grade →Pull it as data →

Run your own AI

Your own AI, on your own computer.

Open models download for free and run on hardware you already own, with nothing sent anywhere. We show what fits your machine, which app to use, and what you give up, without the hype.

Start here: Run your own AI →The hardware guide →Straight answers →

The map

The directory, by what you came to do.

Hand-graded, never scraped. Every shelf is read from the makers' own pages on the day shown, plain links only, nothing for sale.

  • 9 graded · 171 question pages

    Assistants

    Every assistant, every question →
  • 8 graded

    Coding agents

    The shelf, graded →
  • 14 graded

    Memory systems

    The shelf, graded →
  • 12 graded

    Open models

    The shelf, graded →
  • 14 graded

    Inference APIs

    The shelf, graded →
  • 4 categories · 41 tools

    Apps, hardware, hosting, routers

    Every table →
  • The thesis, the framework, the research

    The Continuity Layer

    The Continuity Layer →

Free tools

Every page on the site, in one place →

Join

Your AI will change again next month. The record keeps up, so you don’t have to.

Request occasional research and site updates. Publication follows available capacity. Unsubscribe anytime.

Join the free community ↗YouTube ↗@continuitylayer on X ↗

Questions

Questions people ask first.

What is The Continuity Layer?

An independent record of AI: what changed today with the source, who scores highest on memory, your data, price and leaving, how coding agents, memory systems, open models and inference APIs compare on what you keep, and the state of the field for someone who just arrived. Every grade is read from the maker's own page, dated, and open as data anyone can pull.

The open data

Where do the grades come from?

From each provider's own help and policy pages, read and graded yes, partly or no against a published test, with the day we checked. When we run a test ourselves the row says so. Nothing is scored from a blog or a rumor.

How we grade

Who pays for this?

Nobody. No provider can pay for placement, a test or a result. The site is kept by one independent builder; affiliate links appear only on neutral tools such as hosting and hardware, never on a graded provider, and every one is disclosed.

The disclosure

Can I run AI on my own computer?

Yes. Open models such as Gemma, Qwen and gpt-oss download for free and run on an ordinary laptop with an app like LM Studio or Ollama, with nothing sent anywhere. What fits depends on your memory, and they are a step behind the best hosted models. Our free path takes you from zero to your own setup, with the trade-offs stated plainly.

Run your own AI

What does “continuity layer” mean?

The part of an AI setup you own: a record the next AI reads first and writes to last, so your work survives the AI being replaced. This site runs on one, in the open, and keeps the record for the whole field.

The definition, in full

Contact

Let’s talk about continuity.

Have a question, a correction, or an idea for working together? Send a message to the maintainer.

Your details are used to respond to this message. This form does not subscribe you to updates. Please do not include passwords or other sensitive information.

Disclosure

Built and maintained by one independent builder.

The Continuity Layer is built and maintained by one independent builder who also does commercial work in this field. The System, Governance and Lab pages describe the maintainer's own research prototype; they are project reports, not independent evaluations.

The field map takes no sponsorships and applies the same sourcing and correction standards to the maintainer’s work and outside projects. Inclusion is not endorsement. Source reviews, reported results and independently reproduced evidence are distinguished; commercial interests do not remove the need to disclose limits.