memory
/memory
Recall and add to the Bento workspace's shared memory — what its agents have learned about the user, their people, their decisions and their work — for the whole workspace or one agent, and train it on sources. Use when a task touches past work, people or decisions, or when asked to remember, save, look up, learn or train on something.
- Author
- dzhng ↗
- Repository
- bento-releases ↗
- Licence
- AGPL-3.0
- Last repository change
- 10.11.2026
Assessment and security
- Files in this package
- 1
- Scripts
- 0
- HTML comments
- 0
- URLs in SKILL.md
- 1
No signals detected. Repository changed on 10.11.2026, without references to retired models or APIs.
AGPL-3.0, copyleft: przeróbki udostępniane innym muszą mieć tę samą licencję.
What it does and when to use it
Recall and add to the Bento workspace's shared memory — what its agents have learned about the user, their people, their decisions and their work — for the whole workspace or one agent, and train it on sources. Use when a task touches past work, people or decisions, or when asked to remember, save, look up, learn or train on something.
Original author description
Recall and add to the Bento workspace's shared memory — what its agents have learned about the user, their people, their decisions and their work — for the whole workspace or one agent, and train it on sources. Use when a task touches past work, people or decisions, or when asked to remember, save, look up, learn or train on something.
How it works
- Recall before you answer
- Save when asked
- Train on sources, when asked
SKILL.md preview
---
name: memory
description: Recall and add to the Bento workspace's shared memory — what its agents have learned about the user, their people, their decisions and their work — for the whole workspace or one agent, and train it on sources. Use when a task touches past work, people or decisions, or when asked to remember, save, look up, learn or train on something.
---
# Memory
A Bento workspace remembers. Every session its agents run is observed, condensed
and recalled across sessions, so the workspace holds what the user has told any
of them and what they have learned doing the work. This skill reads and adds to
that shared memory from here. It is the workspace's memory, not this computer's.
Memory is kept at two levels. The workspace's own memory is shared by everyone in
it. Each agent also has its own, which only that agent's sessions recall. Find an
agent's `slug` with `bento agent list`.
## Recall before you answer
When a task turns on something the workspace may already know — a person, a
customer, a past decision, a preference, how something was done last time —
search before answering or asking the user again:
```bash
bento memory search --q "pricing decision for enterprise plans" --limit 10 --json
bento agent memory search --agent-slug sales --q "pricing decision for enterprise plans" --json
```
Search the workspace first, and the agent whose work the task is about when there
is one. Results are ranked by relevance; say what you found and where, and treat
an empty result as "not remembered", not as evidence of the opposite.
## Save when asked
Write to memory only when the user asks you to remember something. Record what
they said in their own words, not your paraphrase:
```bash
bento memory create --content "Prefers metric units in reports" --source user --json
bento agent memory create --agent-slug sales --content "Enterprise discounts cap at 20%" --source user --json
```
Use the workspace for what everyone should know and an agent for what changes
how that one agent works. When it is not clear which, ask. The command answers
with the stored record; report success only once it has.
## Train on sources, when asked
Training reads sources, writes a record of what they say, and installs it in
memory. Start a job only when the user asks to train on something, not because
material came up.
```bash
bento train job create --body-file - <<'JSON'
{"urls":["https://example.com/handbook"]}
JSON
```
`bento agent train job create --agent-slug <slug>` trains one agent instead. Files
already in the Drive train through `driveFiles` paths. To train on a file on this
computer, put it in the Drive first with
`bento file update --path uploads/<name> --in <local file>`.
The job answers with a `jobId`. Poll `bento train job get --job-id <jobId>` (or
`bento agent train job get --agent-slug <slug> --job-id <jobId>`) until it
reaches a terminal state, then tell the user what was learned and where it
lives. `bento train --help` lists the record commands for reviewing or
correcting what training installed.
Versions
| Repository | Score | Licence |
|---|---|---|
| This page · dzhng/bento-releases | 95 | AGPL-3.0 |
| rjmurillo/ai-agents | 100 | MIT |
| rjmurillo/ai-agents | 100 | MIT |
| topoteretes/cognee-integrations | 100 | Apache-2.0 |
| topoteretes/cognee-integrations | 100 | Apache-2.0 |
| topoteretes/cognee-integrations | 100 | Apache-2.0 |
| thebtf/engram | 100 | MIT |
| stxrmik/antigravity-plus | 100 | MIT |
| boshu2/agentops | 100 | Apache-2.0 |
Database snapshot: 13.11.2026 · package version fd21d1b04061 (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 90mnemon
Mnemon is the memory system for this workspace: what the agent knows about the user, the project and past decisions. Use when the user asks what you remember, asks you to remember or forget something, wants to see, search, audit, back up or reconcile stored memory, or before… - 95pact-memory
Persistent memory for PACT agents. Save context, goals, lessons learned, decisions, and entities. Semantic search across sessions. Use when: saving session context, recalling past decisions, searching lessons. Triggers: memory, save memory, search memory, lessons learned, remember, recall - 100brain-memory
Durable memory across conversations, backed by the Brain memory database. Use when storing what the user told you (preferences, decisions, project facts, people and systems), recalling what was said before, correcting something that changed, or reasoning over what is already… - 100using-memory
Use the local memxt memory to recall past decisions, code, and context, and to persist new ones. Invoke whenever the user refers to prior work ("what did we decide", "how did we", "last time", "remember when") or makes a decision worth keeping. - 90team-memory
Gives a team of agents a shared long-term memory that survives across sessions. Use when several agents work for one owner: recall a topic before starting work, save corrections and decisions in full signed with the author's name, let the newest correction win, remove a wrong… - 95manage-agents
Create, configure and inspect agents in a Bento workspace from this computer — register an agent folder, write its instructions and config into the Drive, give it an icon, and change or remove it. Use when setting up an agent persona or repeatable agent process, or changing an existing one.