Repository navigation
PCS 0.9.181
This release brings together the changes since PCS 0.9.180: Local AI replies that no longer wait while PCS stops a memory draft, shorter memory updates on Claude, ChatGPT and OpenRouter, and an In mind right now list that shows which memories are new.
The release includes the Windows x64 application, ready to run offline with its pinned runtime and user guides, and the matching editable source. Tests, developer documents and build tools are only in the Source ZIP.
Local AI replies no longer wait for memory work
- When you spoke or typed while PCS was writing a memory on Local AI, PCS stopped the memory work, but the local model went on writing the stopped draft to its end. In one 0.9.180 session the first reply waited almost 26 seconds before it began.
- PCS's background work on Local AI, including writing and checking memories, Study, and PCS Perspective working in the background, is now streamed, so the local model stops as soon as PCS stops it. In a test on the same PC, a reply that arrived during a long background request waited less than a tenth of a second to start.
- Conversation replies on Local AI are sent as before.
Shorter memory updates on Claude, ChatGPT and OpenRouter
- When new conversation changes a memory that already exists, the memory writer can now send only what changed instead of writing the whole memory again. PCS fills the change into the saved memory before anything checks it, so the memory review still sees the whole memory and the usual protections still apply: an update cannot erase most of a memory, change a memory you corrected yourself, or replace a newer version.
- The writer no longer writes a search description for each memory. PCS always replaced it with its own.
- This rests on one test per version with a made-up conversation on Claude Sonnet 5.5. An update that added one new fact used about half as many output tokens (1,598 instead of 3,265), its draft took 12 seconds instead of 19, and it left 9 of the memory's 11 parts exactly as they were, where 0.9.180 reworded five parts it had no reason to change. An update that changed nearly every part of a memory saved much less (3,589 output tokens instead of 4,075). It has not been tried on a real vault, or on ChatGPT, OpenRouter or Claude Haiku.
- Memory writing on Local AI, Meta, OpenAI and Google is unchanged.
What PCS has in mind (experimental)
- With Show what PCS has in mind turned on, the In mind right now list in the Memory stream panel shows each memory that is still being given to the model. One that came to mind on an earlier turn now reads Still in mind; only one that is new to the latest turn reads Came to mind. Before, the same memories were listed as Came to mind on every turn, so the list looked as if it kept growing. With Let memories come to mind on their own, up to six are kept at a time.
Known limits
- A Local AI reply you interrupt or stop is still written to its end before the next request can start.
- The Local AI change was tested with the recommended Gemma 12B model on an NVIDIA card, not with a custom model or the Vulkan backend, and not yet in a full conversation.
- When the memory writer sends only what changed, a list it changes must be sent whole. If a model sends only part of a list, the missing items are dropped; the memory review and the check against erasing most of a memory are what stand behind this.
- The long memory drafts seen on a real vault in an earlier bug report have not been measured again.
- The known limits of 0.9.180 remain.
AI-generated memories and reviews remain fallible. PCS provides inspection, original sources and correction controls; important interpretations should still be checked.
Upgrading from 0.9.180
- Save your work, choose Quit PCS, and wait for shutdown.
- Back up the entire stopped data folder, including any custom data directory. Copy original Materials separately if stored elsewhere.
- Extract the Windows ZIP into a new writable folder. Follow the included Backup & restore guide to copy or restore your existing data without merging data folders.
- Open Start PCS.exe, unlock with your existing passphrase and check your records. Keep the previous installation and backup while checking the update.
First launch installs the bundled Python runtime and pinned dependencies offline. Optional local AI models, speech packs and inference runtimes remain separate downloads.
Original Materials files are not encrypted by the vault; protect their backups accordingly.
Some newly saved features raise the vault's minimum supported PCS version. Perspective requires 0.9.122; newer native-provider records and Local research require 0.9.126; the newer cloud search index requires 0.9.130; native Claude/ChatGPT memory-thread suggestions require 0.9.134; a saved OpenRouter connection requires 0.9.151; and OpenRouter helper settings and Practice assessments larger than the old 4 KiB limit require 0.9.152. 0.9.155 to 0.9.181 add no new requirement. Opening the vault alone does not activate these requirements. The included vault compatibility guide lists the details.
Support PCS
If PCS is useful to you, you can help support its development. Support is optional.