Codex forgets. Seamless Context remembers.
When Codex compacts a long chat, many models keep almost nothing of what came before: your decisions, the results it found, the files it changed. Seamless Context saves a bounded summary of the whole conversation just before each compaction and gives it back afterwards, once, so your thread picks up where it left off.
One-time payment · 30-day refund · No subscription · Stays on your computer
Our own tests, GPT-5.6 Sol, Light effort, one Windows computer. How we tested and what we have not measured.
Three things happen, automatically
It saves before compaction
A hook runs just before Codex compacts and builds a bounded summary of the whole thread: your goal, your instructions and corrections, what the assistant decided and found, and the commands it ran with short results.
It gives it back once
After the compaction the summary is offered to Codex exactly one time, so it is never replayed or stale. Your thread continues where it left off.
It stays on your computer
Summaries and memory are stored locally. No outgoing connections, no online licence check, and your licence key is stripped from anything it saves.
The summary is built by code from excerpts of your messages, the assistant's replies and the tool results, ranked by how specific they are (numbers, names, paths, decisions). No model rewrites it, so nothing is invented.
It helps most where Codex remembers least
| Model | What we found |
|---|---|
| GPT-5.6 Sol | Large improvement: all 11 facts recalled with it, none without it. |
| GPT-5.6 Terra | Codex already keeps most of a conversation (11 of 11 either way), so there is little to add. |
| GPT-5.6 Luna | Codex already keeps most of a conversation, so there is little to add. |
| GPT-6 models | Not measured yet. |
One price. No subscription.
Introductory price. One-time payment. No subscription.
- 30-day refund, no questions asked
- All 1.x updates included
- Licence for up to 3 of your own computers
- Windows 64-bit and Codex Desktop; no administrator access and no Python needed
- Licence key and personal download link on screen and by e-mail within minutes
By buying you agree to the Terms of Sale and Licence and the Privacy Policy.
Why AI forgets, explained properly
Plain-English guides on context windows, compaction and what actually helps.
Questions people ask
Is it a subscription?
No. One payment covers all 1.x updates.
How do updates work?
Every 1.x update is included at no extra cost. When a new version is released we e-mail it to buyers, and your personal download link always serves the latest version. The customer guide inside the download has the short steps to swap it in.
Which models does it help with?
We measured a large improvement on GPT-5.6 Sol, where Codex on its own keeps almost nothing after a compaction. On GPT-5.6 Terra and GPT-5.6 Luna, Codex already keeps most of a conversation, so there is little for Seamless Context to add. The GPT-6 models have not been measured yet.
Does it send my conversations anywhere?
No. Summaries and memory stay on your computer. The plugin makes no outgoing connections and does not check your licence online.
What exactly is saved?
A bounded summary of the conversation: your goal, your instructions and corrections, short excerpts of what the assistant decided and found, and the commands it ran with short results. It is stored only on your computer and is replaced at the next compaction. Your licence key is stripped from it.
Give your Codex sessions a memory that survives compaction.
US$39 one-time. 30-day refund. Licence key and download link within minutes.
Buy Seamless Context