Skip to content

memory.md

Latest commit

 

History

History
39 lines (32 loc) · 1.92 KB

memory.md

File metadata and controls

39 lines (32 loc) · 1.92 KB
Raw

Persistent Memory

← Back to README

What - a durable, cross-session memory of individual Markdown fact-files catalogued by an auto-maintained MEMORY.md index. Why - the agent otherwise forgets everything between sessions, and re-discovering project facts on every run wastes turns and tokens. How - memory is on by default; the index is injected at session start and the agent reads or writes individual facts through the Memory tool. Configure it in memory.yaml.

# .infer/memory.yaml
enabled: true
dir: ""                # "" => ~/.infer/memory
max_chars: 2000        # cap on the injected MEMORY.md index
max_entry_chars: 2000  # cap on a single fact's content at write time (0 means the default)
backend:
  type: local          # local (default) keeps the directory plain | git backs it with a remote
  git:                 # read only when type is git
    repo: ""
    branch: main
    commit_message: "chore(memory): sync"
    timeout: 60        # seconds per git operation
    sync:
      on_start: pull   # pull (default) | off
      on_finish: push  # push (default) | off

A git-backed memory directory is pulled on run start, and committed and pushed when a fact changes, using the ambient git credential chain (ssh-agent, credential helper, GIT_* environment).

Two default reminders keep memory in use. memory-consult points the agent at the index at session start, and memory-hygiene nudges it once after 25 turns to save what a future session would otherwise miss. Turn memory off with memory.enabled=false (or INFER_MEMORY_ENABLED=false), and both reminders are pruned automatically.

Related