CandleKeep
Manuscripts

A library your agents read — and write back to.

Your agents don't just read knowledge — they write it back. Manuscripts are living books your agents maintain as they work: every session's learnings filed, cited, and waiting for the next agent that needs them.

Task: optimize cold-start latency

LLM Wiki · Index

  • Database connection pooling
  • Rate-limiting the public API
  • Caching strategy

How it works

01

Define it

Set topics, criteria, and writing instructions

02

Agents work

They read and complete your tasks

03

It grows

New, cited entries are written back automatically

What you get

Auto-updating

After each session the agent writes what it learned back into the book; the next agent picks up where the last one left off.

Structured like a wiki

An index page, per-topic pages, and a changelog keep every entry findable and traceable across sessions.

Instructions-as-data

You set the topic scope, capture criteria, and update rules; agents follow them without touching a prompt.

Cited & reversible

Each entry carries its source and date; the changelog makes every change auditable and rollback-ready.

Every session can compound — or reset

Without CandleKeep

Each agent session starts from zero. Hard-won learnings — which API timeout is wrong, which installer trap costs a day — evaporate when the context closes, and the next agent rediscovers them.

With CandleKeep

A learning written back to a Manuscript becomes a chapter every agent reads next time. Knowledge accumulates across sessions instead of inside them.

Knowledge that writes itself.