Skip to main content

atomic vault

Manage the vault (shared project knowledge store).

Synopsis​

atomic vault <SUBCOMMAND>
atomic vault init
atomic vault show <PATH> [--json] [--revision <HASH>]
atomic vault list [-t <TYPE>] [-p <PREFIX>] [--json]
atomic vault materialize [-p <PATH>]
atomic vault sync
atomic vault goal <SUBCOMMAND>
atomic vault context [QUERY]... [--intent <INTENT>] [--files <PATH>]
atomic vault summaries [--goal <GOAL>] [--json]
atomic vault query # alias for `atomic query`

Description​

The vault is Atomic's shared project knowledge store: a structured collection of entries — memories, intents, goals, sessions, skills, scratch notes, and tool results — stored in the repository database and projected to markdown for humans.

Use the vault when knowledge should outlive a single session: durable context for future agent runs, decision records, and task directives. The Knowledge Graph indexes vault content, so vault entries are searchable with atomic query.

Entry Types​

TypePurpose
memoryDurable context (decisions, lessons, constraints, preferences) — author with atomic memory
intentDirective-based units of work (a :::why, acceptance criteria, tasks) — author with atomic intent
goalA started unit of work being tracked with tool-result summaries
sessionAgent session records
skillReusable instructions
scratch / tool_resultWorking notes and captured tool output
note

atomic vault intent and atomic vault memory were removed; they now print a redirect telling you to use atomic intent and atomic memory instead.

Subcommands​

vault init — Initialize the Vault​

Initialize the vault in an existing Atomic repository.

atomic vault init

vault show — Print an Entry's Content​

Synopsis​

atomic vault show <PATH> [OPTIONS]

Arguments​

<PATH> — Vault-relative path to the entry (e.g. memory/architecture.md).

Options​

OptionDescription
--jsonOutput as JSON instead of markdown
--revision <HASH>Require the entry to match this exact revision before printing JSON
atomic vault show memory/architecture.md

vault list — List Entries​

Synopsis​

atomic vault list [OPTIONS]

Options​

OptionDescription
-t, --type <TYPE>Filter by entry type (session, memory, intent, skill, scratch, tool_result)
-p, --prefix <PREFIX>Filter by path prefix
--jsonOutput as JSON
atomic vault list --type memory
atomic vault list --prefix intent/ --json

vault materialize — Write Entries to Disk​

Materialize vault entries to markdown files on disk.

Synopsis​

atomic vault materialize [OPTIONS]

Options​

OptionDescription
-p, --path <PATH>Materialize only this specific vault path
atomic vault materialize                 # everything
atomic vault materialize -p memory/ # one subtree

vault sync — Sync Markdown Back to the Database​

Sync vault markdown files (as materialized on disk) back into the vault database.

atomic vault sync

vault goal — Manage Goals​

Track a unit of work end-to-end: start it, capture tool-result summaries, stop/resume it.

Synopsis​

atomic vault goal <SUBCOMMAND>

Subcommands​

goal start — Start a new goal.

OptionDescription
--name <NAME>Override the generated goal name
--developer <DEVELOPER>Developer name
--intent <INTENT>Link to an intent ID (e.g. PIMO-1 — see atomic intent list)
--model <MODEL>AI model being used
--jsonOutput as JSON

goal stop [GOAL] — Stop the current (or named) goal. --promote marks it completed (promoted for team consumption); --discard deletes it entirely.

goal resume <GOAL> — Resume a suspended goal (--json for JSON output).

goal list — List goals. -s, --status <STATUS> filters by active, completed, suspended, or all (default). --json for JSON.

goal show <GOAL> — Show a goal's content (--json for JSON).

atomic vault goal start --intent PIMO-1 --model glm-4
atomic vault goal list --status active
atomic vault goal stop --promote

vault context — Research Relevant Memories​

Find vault memories relevant to a task. This is the command agent integrations call to load durable context before working.

Synopsis​

atomic vault context [QUERY]... [OPTIONS]

Arguments​

[QUERY]... — Free-text query terms.

Options​

OptionDescription
--intent <INTENT>Seed from an intent: uses its title, labels, and body as query terms, plus its graph neighbors as candidates (e.g. PIMO-1)
--files <PATH>Seed from a file path; memories referencing it are candidates
--limit <LIMIT>Maximum memories to return (default: 5)
--budget-chars <BUDGET_CHARS>Total character budget across all memory bodies (default: 8000)
--candidates-onlyReturn ranked candidate metadata without memory bodies
--format <FORMAT>md (prompt-ready block) or json (default: md)
--jsonOutput as JSON (shorthand for --format json)
atomic vault context --intent PIMO-1
atomic vault context "authentication tokens"
atomic vault context --files src/auth/login.rs --json
atomic vault context "authentication" --candidates-only --json

vault summaries — Get Tool Result Summaries​

Get tool result summaries for a goal. --goal <GOAL> names the goal (see atomic vault goal list); --json for JSON output (JSON is the default for summaries).

atomic vault summaries --goal fix-token-refresh --json

vault query — Query the Knowledge Graph​

Alias for atomic query.

See Also​

  • intent — Authoring directive-based intents
  • memory — Authoring durable memories
  • query — Searching the knowledge graph