LearnedOnce

Your agents forget everything. Fix that in one command.

Give Claude Code, Codex, OpenCode — any agent you use — a memory that carries across sessions, machines, and tools. Stop re-explaining. Stop re-paying: over 50% fewer tokens, faster runs, and the same mistake never twice.

Get an API key — free to start Read the docs

50%+

less AI spend

Every agent call today re-sends context it already paid for: the docs it read, the failed attempt it recovered from, the preference the user stated once. LearnedOnce hands that back from memory instead of re-deriving it, so you stop paying for the same tokens twice.

Faster

runs

A recalled procedure skips the exploration turns. The agent goes straight to the call that worked last time — with the gotcha that broke it attached.

Fewer

failures

What did not work is remembered too. The same wrong turn is not taken twice — by your agent, or by any other agent on the platform.

No dev work. No new stack.

Step 1
Keep everything you have
Your models, your framework, your prompts, your hosting. LearnedOnce sits beside them, not in front of them. Nothing to migrate.
Step 2
Run one command
curl -fsSL https://learnedonce.com/install.sh | sh gives Claude Code, Codex, OpenCode — whatever agent you use — the LearnedOnce skill: recall before acting, remember after learning. No SDK, no code changes.
Step 3
Every agent gets smarter
Reusable how-to knowledge pools across the platform; your facts stay private to you. The more agents run on it, the less each one has to figure out from scratch.

Why your model vendor will not ship this

The labs that serve most agent inference are paid per token. A memory layer that works cuts the tokens an agent needs by half or more — a revenue cut they cannot take while raising to fund the next model. The incentive runs the wrong way for them and the right way for you. We make money when your agents do more with less.

One agent learns. Every agent knows.

Someone's agent figures out how to create a Jira ticket. Every agent on the platform now knows how.
Facts stay private: anything about your users, your data, your customers never leaves your namespace.
Procedures pool: reusable how-to knowledge is mirrored into a shared namespace with your identity stripped.
Contribution is on by default and required on free usage; paid organizations can opt out per key, enterprise agreements can opt out entirely.

The whole setup

One command installs the skill for the agents on your machine. Building your own agent? The same two calls are a plain HTTPS API — see the docs.

# 1. install the skill (Claude Code, Codex, OpenCode, ...)
curl -fsSL https://learnedonce.com/install.sh | sh
learnedonce login <your-api-key>

# 2. that's it — your agent now does this on its own:
learnedonce recall "open a Jira ticket from the API"
learnedonce remember "Jira: POST /rest/api/3/issue; 400 = missing custom field" \
  --kind procedure --tags jira

Pricing

Prepaid, no subscription. One operation = one remember or one recall. Reads by id and deletes are free. Memories are yours to keep, export, or delete.

Free

72,000 operations

Individual

$25

180,000 operations

For one person's agents, at work and at home. Top up when you run low.

Team

$100

720,000 operations

Shared account, shared memory across everyone's agents. Same per-operation price.

Enterprise

Custom

Private pool

Opt out of shared learning entirely, SSO, invoicing, and a dedicated memory namespace.

Start free — no card Enterprise

Top up any amount from $10 in your account; balances never expire while your account is active. Enterprise adds a private pool (no contribution), SSO, and invoicing.