Glossary¶

The words the documentation uses, each with one meaning.

Agent¶

A program that works for you with a language model: Claude Code, Codex, OpenCode, Hermes Agent, Antigravity. Agents produce sessions and read Jezero.

App¶

The terminal app for people that jezero opens. See Use the Jezero app.

Atom¶

One fact worth keeping, extracted from a session: a decision, its rationale, a fix, an exact error, a command that worked, a fact or a risk. An item holds several. See Data model.

Client¶

Anything allowed to read the library: one [[clients]] block in jezero.toml, with one key. A device is a client created by the plugins. See Configuration.

Client core¶

jezero_client.py, one Python file with no dependencies that every plugin carries. It creates the device's key, runs the hooks, queues uploads and talks to the server. See Client reference.

Connector¶

The part of the server that reads one source (an agent's session files, a task board) and turns new material into documents to summarise. Each keeps a cursor so it only reads what is new. See Connectors.

Device¶

A computer connected with the plugins. It has its own key, made on that computer; the server only stores the key's SHA-256 fingerprint. A device either uploads its sessions or only reads. See Manage computers and keys.

Hook¶

A command an agent runs at a moment of its life (session start, first message, end). Jezero's hooks add context at the start and upload the session at the end. They never block or break the agent.

Inbox¶

The folder on the server where devices upload raw session files. Connectors read it like any local store. See HTTP API.

Item¶

One entry of the library: a session, or a part of a long one, a task or a thread, summarised. It has a title, a summary, atoms, the questions it answers, a project, a source and a date. The app calls items memories. See Data model.

Key¶

The secret a client presents to the server, as Authorization: Bearer ***. Plugins keep it in the device's client home and never print it.

Library¶

Everything Jezero has stored: one SQLite file with full-text search.

MCP¶

The Model Context Protocol, the standard way agents call tools. Jezero's MCP server is read-only. See MCP tools.

Plugin¶

The package that connects one agent to Jezero: the MCP server, the hooks and the skill. See Connect an agent.

Project¶

What an item is about (atlas, billing). Jezero infers it from the session's folder or the summary, and you can map folders to projects by hand. See Add and tune sources.

Protocol¶

The short text the session-start hook gives every agent: search Jezero by habit, check with later before relying on something, cite item ids.

Read-only computer¶

A device whose key may search but not upload, for computers whose sessions already reach Jezero another way, or should not.

Source¶

Where an item comes from: the name of its connector, such as agent-codex-laptop. Keys are limited to the sources they may read.

Summarizer¶

The language model that writes each item from the original text. A small, fast model is enough. See How summaries are made.

Trust¶

own for your agents' work, external for text written by others. Agents see it on every result.