Three-Tier Agent Memory

Memory that survives restarts without filling the context window, in plain files any agent can read.

Three-Tier Agent Memory product illustration

The three tiers

  • Working: memory/NOW.md, current goal, next step, blockers (loaded every session, max 40 lines)
  • Episodic: memory/log/YYYY-MM-DD.md, one dated line per event (searched, never loaded whole)
  • Long-term: memory/MEMORY.md, durable facts with date and source (loaded every session, max 150 lines)

The script

memory.py does init, log, promote (dedupes, refuses secrets), recall (ranked keyword search across all three tiers), lint, forget and archive. 19 tests; Python 3.9+, standard library only.

Honest limit

Recall is keyword search, not embeddings: write facts in the words you will search for.

FAQ

Does it use a vector database?
No, plain Markdown and keyword recall.
Where is memory stored?
In a memory/ folder in your project.
Can it leak secrets?
It refuses to store common key, password and card patterns, and lint flags any it finds.
How do I get the files?
Right after payment, Stripe’s confirmation page shows your download link.
Refunds?
Email hello@youraitools.co within 14 days if it doesn’t work and we can’t fix it.
Can I pay with crypto?
Yes. Pay the same price in USDC or USDT on Base, Ethereum, Arbitrum, Optimism, Polygon or Solana from the "Pay with crypto" button, with no account or KYC. We check the payment on-chain and show your download links as soon as it confirms.

Get Three-Tier Agent Memory

Agent memory in plain files: working note, daily log, long-term facts. No vector DB.

Save with a bundle

Pairs well with

Download$7

Autonomy Ladder

Human-in-the-loop policy: per action, your agent acts, asks or never acts.

View Autonomy Ladder

Related