Recall

All-day sessions. No compaction.

The model gets your recent work and an index of the rest. Older details are looked up word for word, not summarized away.

Start freeRead the docs
Try it
Your agent sendssystem prompttoolsthe full history
The model getssystem prompttoolsmemory blockindex of older workrecent work, unchanged

Older work goes to the archive, word for word.

⚡ context-mode · picked up after a 2 h pause, nothing compacted

After a long pause, the prompt is rebuilt shorter.

⚡ context-mode · recalled 2 items from 2 days ago (a decision, src/billing/quota.ts)

The agent reads older work back when it needs it.

How it works

Nothing summarized. Nothing lost.

  1. 01Your agent sends everything

    As usual. Your client keeps its full history.

  2. 02Older work is archived

    Word for word, found by id, words or time.

  3. 03The model looks it up

    When it needs an early detail.

Proof

One long session, run twice.

Cost per session, 200K window

Direct$4.86
Context Mode$3.04

37.6% lower

Cost per session, 1M window

Direct$7.24
Context Mode$2.97

59.0% lower

Early details answered exactly

Direct4 of 10
Context Mode9 of 10

5 more

Compactions

Direct3
Context Mode0

none

Pre-registered paired runs, Claude Opus 5.5, 2026-10-03 and 2026-10-04. Data: 200K window, 1M window. Method and sources →

Features

Made for long sessions.

Kept 90 days

By default. Forget a project to delete it.

Lookups you can see

The console lists each read.

Sub-agents too

A line a sub-agent read can be found later.

One session, all day.

npx @context-mode/cli connects your agents. 1,000 requests free, no card.

Start free →Read the docs →