Question 1

What does this extension actually do for long coding sessions?

Accepted Answer

It keeps long sessions from losing evidence: it cuts history into chunks, asks Jev keep/kind questions about each candidate, stores the surviving original lines in a verbatim ledger, and renders compaction deterministically. The transcript itself is never rewritten.

Question 2

How is Jev used inside the extension?

Accepted Answer

Only for two decision types: observe (keep this candidate? what kind is it?) and consolidate (promote it into a durable topic file?). Jev answers typed questions rather than generating text, and the extension calls https://api.typesafe.ai/v1/systemone directly.

Question 3

How do I install it?

Accepted Answer

Create a TypeSafe account and API key at typesafe.ai, export TYPESAFE_API_KEY in the environment that launches pi, then run pi install git:github.com/willfish/pi-observational-memory-jev. For local development the README shows pi -e ./src/index.ts.

Question 4

Is the extension enabled by default?

Accepted Answer

No. The package default is off; turn it on per session with /om on, or set enabledByDefault in settings. State persists in the ledger as om.enabled and survives resume, and PI_OM_PASSIVE=1 disables clocks without flipping the on/off gate.

Question 5

What commands and gauges does it add?

Accepted Answer

/om, /om on and /om off gate the feature; /om:status shows workers, pool, clocks, the Jev endpoint and the last error; /om:compact and /om:consolidate force work. Footer gauges show observer progress, the consolidator pool, context and session cost.

Question 6

Can it run alongside Alvar's original /om extension?

Accepted Answer

No: the README says the commands collide, so pick one. This project follows that extension's slash commands, clocks, ledger, footer gauges and .memory/<session>/ layout, but swaps the chat-model pi subprocess observer for TypeSafe Jev.

Question 7

What does it cost and what are the known limits?

Accepted Answer

It needs a TypeSafe API key, and Jev pricing is per input token per the linked pricing page; the captured README publishes no cost or latency figures of its own. The excerpt also cuts off inside the example output, so treat the documented settings as the source of truth.

Pi Observational Memory Jev

Highlights

Quickstart

Questions about this entry

From the Jev guides

From the Jev AI model community