Enum Command 

pub enum Command {
    Serve(ServeArgs),
    Mcp {
        tier: String,
        profile: Option<String>,
    },
    Store(StoreArgs),
    Update(UpdateArgs),
    Recall(RecallArgs),
    Search(SearchArgs),
    Get(GetArgs),
    List(ListArgs),
    Delete(DeleteArgs),
    Promote(PromoteArgs),
    Forget(ForgetArgs),
    Link(LinkArgs),
    Consolidate(ConsolidateArgs),
    Gc,
    Stats,
    Namespaces,
    Namespace(NamespaceArgs),
    Config(ConfigCliArgs),
    Export,
    Import(ImportArgs),
    Resolve(ResolveArgs),
    Shell,
    Sync(SyncArgs),
    SyncDaemon(SyncDaemonArgs),
    AutoConsolidate(AutoConsolidateArgs),
    Completions(CompletionsArgs),
    Man,
    Mine(MineArgs),
    Archive(ArchiveArgs),
    Agents(AgentsArgs),
    Identity(IdentityArgs),
    Offload(OffloadArgs),
    Deref(DerefArgs),
    Rules(RulesArgs),
    Pending(PendingArgs),
    Backup(BackupArgs),
    Restore(RestoreArgs),
    Curator(CuratorArgs),
    Bench(BenchArgs),
    Doctor(DoctorCliArgs),
    Boot(BootArgs),
    Install(InstallArgs),
    Wrap(WrapArgs),
    Logs(LogsArgs),
    Audit(AuditArgs),
    Governance(GovernanceCliArgs),
    VerifyReflectionChain(VerifyChainArgs),
    VerifySignedEventsChain(VerifySignedEventsChainArgs),
    ExportForensicBundle(ExportForensicBundleArgs),
    VerifyForensicBundle(VerifyForensicBundleArgs),
    ExportReflections(ExportReflectionsArgs),
    RecoverPreviousSession(RecoverPreviousSessionArgs),
    Atomise(AtomiseArgs),
    Persona(PersonaArgs),
    Calibrate(CalibrateArgs),
    Skill(SkillArgs),
    Share(ShareArgs),
    KgQuery(KgQueryArgs),
    FindPaths(FindPathsArgs),
    RecallObservations(RecallObservationsArgs),
    Expand(ExpandArgs),
    CheckDuplicate(CheckDuplicateArgs),
    Reembed(ReembedArgs),
    Replay(ReplayArgs),
    Reflect(ReflectArgs),
    Subscribe(SubscribeArgs),
    Unsubscribe(UnsubscribeArgs),
    ListSubscriptions(ListSubscriptionsArgs),
    SubscriptionReplay(SubscriptionReplayArgs),
    SubscriptionDlqList(SubscriptionDlqListArgs),
    Notify(NotifyArgs),
    Inbox(InboxArgs),
    IngestMultistep(IngestMultistepArgs),
    KgInvalidate(KgInvalidateArgs),
    KgTimeline(KgTimelineArgs),
    EntityRegister(EntityRegisterArgs),
    EntityGetByAlias(EntityGetByAliasArgs),
    DependentsOfInvalidated(DependentsOfInvalidatedArgs),
    ReflectionOrigin(ReflectionOriginArgs),
    QuotaStatus(QuotaStatusArgs),
}

Variants

Serve(ServeArgs)

Start the HTTP memory daemon.

Tier resolution. Unlike mcp / store / recall, the serve subcommand does NOT accept a --tier flag. The daemon’s effective feature tier is resolved from the tier field in config.toml, falling back to the compiled-in default (semantic). For per-invocation tier overrides use the mcp / store / recall subcommands, which expose --tier directly. See docs/ADMIN_GUIDE.md §“Feature tiers” and issue #703 for the rationale (a long-running daemon owns embedder / LLM resources that are expensive to swap mid-run, so tier is fixed at startup via configuration).

Mcp

Run as an MCP (Model Context Protocol) tool server over stdio

Fields

tier: String

Feature tier: keyword (FTS only) or semantic (embeddings + FTS)

profile: Option<String>

v0.6.4 — Tool surface profile. One of core, graph, admin, power, full, or a comma-separated custom list (e.g., core,graph,archive). Default core (7 tools at v0.7.0: the original 5 + memory_load_family + memory_smart_load). Resolution order: this CLI flag > AI_MEMORY_PROFILE env > [mcp].profile in config.toml > core. Set --profile full to expose every family — at v0.7.0, Profile::full().expected_tool_count() returns 74 (canonical SSOT; pinned by profile_full_matches_registry_all against crate::mcp::registry::tool_names::ALL.len()). The 74 advertised entries decompose as 73 callable “memory tools” plus the always-on memory_capabilities bootstrap; the build_capabilities_summary “{n} memory tools” phrasing reports the 73 memory-tool count to reconcile with the user-facing summary (see issue #862 for the disambiguation).

Store(StoreArgs)

Store a new memory

Update(UpdateArgs)

Update an existing memory by ID

Recall(RecallArgs)

Recall memories relevant to a context

Search(SearchArgs)

Search memories by text

Get(GetArgs)

Retrieve a memory by ID

List(ListArgs)

List memories

Delete(DeleteArgs)

Delete a memory by ID

Promote(PromoteArgs)

Promote a memory to long-term

Forget(ForgetArgs)

Delete memories matching a pattern

Link(LinkArgs)

Link two memories

Consolidate(ConsolidateArgs)

Consolidate multiple memories into one

Gc

Run garbage collection

Stats

Show statistics

Namespaces

List all namespaces

Namespace(NamespaceArgs)

v0.7.0 (issue #800) — operator CRUD for the per-namespace standard policy memory pointer (Batman Mode Crack 1). Three verbs: set-standard / get-standard / clear-standard, plus the batman-policy helper that prints the canonical Batman GovernancePolicy JSON blob. Closes the friction that kept Batman Forms 2 + 6 dormant on most installs by replacing the MCP-stdio JSON-RPC dance with first-class CLI surface.

Config(ConfigCliArgs)

v0.7.x (#1146) — enterprise configuration tooling. ai-memory config migrate rewrites a legacy v1 (flat-field) config.toml to the v2 sectioned shape ([llm], [embeddings], [reranker], [storage]) with a timestamped .bak backup. --dry-run prints the diff without writing. --also-clean-claude-json additionally removes the mcpServers.<*>.env block from ~/.claude.json after the operator has verified the new config.

Export

Export all memories as JSON

Import(ImportArgs)

Import memories from JSON (stdin)

Resolve(ResolveArgs)

Resolve a contradiction — mark one memory as superseding another

Shell

Interactive memory shell (REPL)

Sync(SyncArgs)

Sync memories between two database files

SyncDaemon(SyncDaemonArgs)

Run the peer-to-peer sync daemon — continuously exchange memories with one or more HTTP peers (Phase 3 Task 3b.1). The defining grand-slam capability: two agents on two machines form a live knowledge mesh with no cloud, no login, no SaaS.

AutoConsolidate(AutoConsolidateArgs)

Auto-consolidate short-term memories by namespace

Completions(CompletionsArgs)

Generate shell completions

Man

Generate man page

Mine(MineArgs)

Import memories from historical conversations (Claude, ChatGPT, Slack exports)

Archive(ArchiveArgs)

Manage the memory archive (list, restore, purge, stats)

Agents(AgentsArgs)

Register or list agents (Task 1.3)

Identity(IdentityArgs)

v0.7 (Track H, Task H1) — per-agent Ed25519 keypair lifecycle. generate / import / list / export-pub against the local key directory (default <config>/ai-memory/keys). Hardware-backed key storage (TPM/HSM/Secure Enclave) is out of OSS scope and lives in the AgenticMem commercial layer.

Offload(OffloadArgs)

v0.7.0 QW-3 — context-offload substrate primitive. Persists a file (or - for stdin) into the offloaded_blobs substrate and prints the short ref_id callers keep in their working window. Pairs with ai-memory deref <ref_id>.

Deref(DerefArgs)

v0.7.0 QW-3 — dereference a previously-offloaded ref_id. Refuses tampered rows (SHA-256 mismatch). Pairs with ai-memory offload <file>.

Rules(RulesArgs)

v0.7.0 (issue #691) — substrate-level agent-action rules engine. CRUD over the governance_rules table consulted by check_agent_action. Mutation verbs (add/enable/disable/remove) require the operator’s Ed25519 keypair on disk at <key-dir>/operator.priv (mode 0600); without --sign they refuse with governance.no_operator_key. Read verbs (list / check) are unprivileged.

Pending(PendingArgs)

List / approve / reject governance-pending actions (Task 1.9)

Backup(BackupArgs)

v0.6.0.0: snapshot the SQLite database to a timestamped backup file. Uses SQLite VACUUM INTO which is hot-backup safe (no daemon stop required). Writes a manifest.json alongside (sha256 + version).

Restore(RestoreArgs)

v0.6.0.0: restore the SQLite database from a backup file written by ai-memory backup. Verifies the manifest sha256 before replacing the current DB. The current DB is moved aside as a safety net before the replacement.

Curator(CuratorArgs)

v0.6.1: run the autonomous curator. --once runs a single sweep and prints a JSON report; --daemon loops with --interval-secs between cycles. Auto-tags memories without tags and flags contradictions against nearby siblings in the same namespace.

Bench(BenchArgs)

v0.6.3 (Pillar 3 / Stream E): run the canonical performance workload and print measured p50/p95/p99 against the budgets in PERFORMANCE.md. Each invocation seeds a disposable temp DB so the user’s main DB is untouched. Exits non-zero when any p95 exceeds its budget by more than the published 10% tolerance.

Doctor(DoctorCliArgs)

v0.6.3.1 (P7 / R7): operator-visible health dashboard. Reads Capabilities v2 (P1) + data integrity surfaces (P2) + recall observability (P3). With --remote <url> becomes a fleet doctor at T3+. Read-only — never mutates the database. Exits 0 on a healthy report, 2 on critical findings, and 1 on warnings when --fail-on-warn is passed.

Boot(BootArgs)

Issue #487: emit session-boot context. Universal primitive every AI-agent integration recipe (Claude Code SessionStart hook, Cursor / Cline / Continue / Windsurf system-message, Codex / Apps SDK / Agent SDK programmatic prepend, OpenClaw built-in, local models via LM Studio / Ollama / vLLM) calls before the agent’s first turn. Read-only, fast, never blocks. With --quiet (recommended for hooks) a missing DB exits 0 with empty stdout.

Install(InstallArgs)

Issue #487 PR-2: wire ai-memory boot and the ai-memory-mcp server into AI agents’ config files (Claude Code SessionStart hook, Cursor / Cline / Continue / Windsurf / OpenClaw MCP config). Default is --dry-run (prints the diff, writes nothing). Pass --apply to commit. Pass --uninstall --apply to remove a previously-installed managed block.

Wrap(WrapArgs)

Issue #487 PR-6: cross-platform Rust replacement for the bash / PowerShell wrappers PR-1 shipped in the integration recipes. Runs ai-memory boot in-process, builds a system message, then spawns the named agent CLI with the system message delivered via the strategy chosen by default_strategy(<agent>) (or an explicit --system-flag / --system-env / --message-file-flag override). Exit code is propagated from the wrapped agent.

Logs(LogsArgs)

Issue #487 PR-5: operator-facing CLI for the operational logging facility (tail, cat, archive, purge). Default-OFF — emits nothing useful unless [logging] enabled = true is set in config.toml.

Audit(AuditArgs)

Issue #487 PR-5: operator-facing CLI for the security audit trail (verify, tail, path). Default-OFF — emits nothing useful unless [audit] enabled = true is set in config.toml.

Governance(GovernanceCliArgs)

v0.7.0 K11 — translate legacy [governance] policies in config.toml into the v0.7 [[permissions.rules]] (K9) format. Default mode is dry-run: prints to stdout. Pass --config-out PATH to write the rendered block to a file (or merge in-place when PATH matches the loaded config).

VerifyReflectionChain(VerifyChainArgs)

v0.7.0 L1-3 — external verifier for reflection chains (procurement-grade audit tool). Walks reflects_on edges backward from <memory_id> to depth 0, verifies each Ed25519 signature, and emits a structured chain-integrity report. Exit 0 if fully verified; non-zero otherwise.

VerifySignedEventsChain(VerifySignedEventsChainArgs)

v0.7.0 V-4 closeout (#698) — walk the SQL-side signed_events cross-row hash chain (schema v34) and emit a structured report. Distinct from verify-reflection-chain (which walks reflects_on edges) and from audit verify (which walks the JSONL audit log). Exit 0 if the chain holds; 1 on chain break.

ExportForensicBundle(ExportForensicBundleArgs)

v0.7.0 L2-5 (issue #670) — export a procurement-grade forensic evidence bundle (signed tarball) for a memory and its reflection chain. The OSS surface for the AgenticMem Attest tier; see crate::forensic::bundle for the bundle layout.

VerifyForensicBundle(VerifyForensicBundleArgs)

v0.7.0 L2-5 (issue #670) — verify a forensic evidence bundle. Re-hashes every file, checks the manifest signature when present, and re-verifies every edge signature against the bundled observed_by public key.

ExportReflections(ExportReflectionsArgs)

v0.7.0 QW-1 — write every reflection memory to a file under ~/.ai-memory/reflections/<namespace>/<id>.md (or .json with --format json) so operators can cat what the substrate has synthesised without learning SQL. The on-disk artefact is derived; the SQL row stays canonical.

RecoverPreviousSession(RecoverPreviousSessionArgs)

v0.7.0 (issue #1389) — fail-safe recovery of agent context from a host’s per-turn transcript file when the previous session terminated ungracefully (SIGKILL, tmux lockup, host crash) between turns. Closes the #1388 substrate failure mode. Designed for SessionStart-hook chaining after ai-memory boot; the in-session counterpart is the memory_recover_previous_session MCP tool.

Atomise(AtomiseArgs)

v0.7.0 WT-1-F — operator-side wrapper over the atomisation engine (crate::atomisation::Atomiser). Decomposes one long-form memory into atomic propositions; surfaces every substrate failure with a stable exit code (see crate::cli::commands::atomise::exit_code).

Persona(PersonaArgs)

v0.7.0 QW-2 — fetch (or regenerate) the Persona artefact for an entity. Read-only by default; pass --regenerate to run the curator and persist a fresh row.

Calibrate(CalibrateArgs)

v0.7.0 Form 5 (issue #758) — calibration driver verbs. ai-memory calibrate confidence --from-shadow reads confidence_shadow_observations and emits per-(namespace, source) baselines computed over the window.

Skill(SkillArgs)

v0.7.0 Cluster E API-2 (issue #767) — ai-memory skill <register|list|get|resource|export|promote|compose> CLI parity surface for the 7 L1-5 Agent Skills MCP tools. Dispatches into the same substrate handlers (re-exported under crate::mcp::handle_skill_*); no business logic is duplicated.

Share(ShareArgs)

v0.7.0 #1095 — ai-memory share subcommand. Closes the SR-4 three-surface-parity gap. Copies a memory into the recipient agent’s shared namespace _shared/<from>→<to>/ via the same substrate primitive the MCP tool (memory_share) and HTTP route (POST /api/v1/share) consume — guaranteeing byte-equal envelopes across the three surfaces.

KgQuery(KgQueryArgs)

v0.7.0 ARCH-3 / FX-12 — ai-memory kg-query subcommand. Outbound KG traversal from a source memory (<=5 hops). CLI parity for the MCP memory_kg_query tool.

FindPaths(FindPathsArgs)

v0.7.0 ARCH-3 / FX-12 — ai-memory find-paths subcommand. Enumerate up to N paths through the KG between two memories (BFS, max_depth<=7). CLI parity for memory_find_paths.

RecallObservations(RecallObservationsArgs)

v0.7.0 ARCH-3 / FX-12 — ai-memory recall-observations subcommand. List rows from the recall-consumption ledger (#886). CLI parity for memory_recall_observations.

Expand(ExpandArgs)

v0.7.0 #1443 — ai-memory expand subcommand. LLM query-expansion over a free-text query. CLI parity for the MCP memory_expand_query tool + the POST /api/v1/expand_query HTTP route — all three share crate::mcp::handle_expand_query. Lets a harness inject expansion as a one-shot without an MCP stdio server or HTTP daemon. Requires a configured LLM (any tier via AI_MEMORY_LLM_BACKEND, or smart/autonomous preset).

CheckDuplicate(CheckDuplicateArgs)

v0.7.0 ARCH-3 / FX-12 — ai-memory check-duplicate subcommand. Pre-write near-duplicate check via cosine over stored embeddings. CLI parity for memory_check_duplicate. Requires the embedder (semantic tier or above).

Reembed(ReembedArgs)

v0.7.0 #1598 — ai-memory reembed subcommand. Full-corpus vector-space migration: re-embeds every live memory (optionally --namespace-filtered) with the resolved embedding backend/model and REPLACES the stored vectors (unlike the boot backfill, which only fills missing ones). --dry-run prints the plan; per-row #1595 failure isolation (skip-with-WARN) keeps one poison row from stopping the sweep. Resolves the embedder via the same AppConfig::resolve_embeddings() + Embedder::from_resolved path as daemon/MCP boot.

Replay(ReplayArgs)

v0.7.0 ARCH-3 / FX-12 — ai-memory replay subcommand. Reconstruct the conversation transcript chain that produced a memory. CLI parity for memory_replay.

Reflect(ReflectArgs)

v0.7.0 ARCH-3 / FX-C3 (batch2) — ai-memory reflect. CLI parity for memory_reflect. CLI dispatcher uses active_keypair=None / embedder=None; operators who need signing or LLM dedup drive the daemon via MCP / HTTP.

Subscribe(SubscribeArgs)

v0.7.0 ARCH-3 / FX-C3 (batch2) — ai-memory subscribe. CLI parity for memory_subscribe.

Unsubscribe(UnsubscribeArgs)

v0.7.0 ARCH-3 / FX-C3 (batch2) — ai-memory unsubscribe. CLI parity for memory_unsubscribe.

ListSubscriptions(ListSubscriptionsArgs)

v0.7.0 ARCH-3 / FX-C3 (batch2) — ai-memory list-subscriptions. CLI parity for memory_list_subscriptions.

SubscriptionReplay(SubscriptionReplayArgs)

v0.7.0 ARCH-3 / FX-C3 (batch2) — ai-memory subscription-replay. CLI parity for memory_subscription_replay.

SubscriptionDlqList(SubscriptionDlqListArgs)

v0.7.0 ARCH-3 / FX-C3 (batch2) — ai-memory subscription-dlq-list. CLI parity for memory_subscription_dlq_list.

Notify(NotifyArgs)

v0.7.0 ARCH-3 / FX-C3 (batch2) — ai-memory notify. CLI parity for memory_notify.

Inbox(InboxArgs)

v0.7.0 ARCH-3 / FX-C3 (batch2) — ai-memory inbox. CLI parity for memory_inbox.

IngestMultistep(IngestMultistepArgs)

v0.7.0 ARCH-3 / FX-C3 (batch2) — ai-memory ingest-multistep. CLI parity for memory_ingest_multistep. CLI dispatcher passes handler=None; tier-locked advisory returns on every tier because the CLI does not own the LLM dispatch.

KgInvalidate(KgInvalidateArgs)

v0.7.0 ARCH-3 / FX-C3 (batch2) — ai-memory kg-invalidate. CLI parity for memory_kg_invalidate.

KgTimeline(KgTimelineArgs)

v0.7.0 ARCH-3 / FX-C3 (batch2) — ai-memory kg-timeline. CLI parity for memory_kg_timeline.

EntityRegister(EntityRegisterArgs)

v0.7.0 ARCH-3 / FX-C3 (batch2) — ai-memory entity-register. CLI parity for memory_entity_register.

EntityGetByAlias(EntityGetByAliasArgs)

v0.7.0 ARCH-3 / FX-C3 (batch2) — ai-memory entity-get-by-alias. CLI parity for memory_entity_get_by_alias.

DependentsOfInvalidated(DependentsOfInvalidatedArgs)

v0.7.0 ARCH-3 / FX-C3 (batch2) — ai-memory dependents-of-invalidated. CLI parity for memory_dependents_of_invalidated.

ReflectionOrigin(ReflectionOriginArgs)

v0.7.0 ARCH-3 / FX-C3 (batch2) — ai-memory reflection-origin. CLI parity for memory_reflection_origin.

QuotaStatus(QuotaStatusArgs)

v0.7.0 ARCH-3 / FX-C3 (batch2) — ai-memory quota-status. CLI parity for memory_quota_status.

Trait Implementations

impl FromArgMatches for Command

fn from_arg_matches(__clap_arg_matches: &ArgMatches) -> Result<Self, Error>

Instantiate Self from ArgMatches, parsing the arguments as needed.

fn from_arg_matches_mut( __clap_arg_matches: &mut ArgMatches, ) -> Result<Self, Error>

Instantiate Self from ArgMatches, parsing the arguments as needed.

fn update_from_arg_matches( &mut self, __clap_arg_matches: &ArgMatches, ) -> Result<(), Error>

Assign values from ArgMatches to self.

fn update_from_arg_matches_mut<'b>( &mut self, __clap_arg_matches: &mut ArgMatches, ) -> Result<(), Error>

Assign values from ArgMatches to self.

impl Subcommand for Command

fn augment_subcommands<'b>(__clap_app: Command) -> Command

Append to Command so it can instantiate Self via FromArgMatches::from_arg_matches_mut

fn augment_subcommands_for_update<'b>(__clap_app: Command) -> Command

Append to Command so it can instantiate self via FromArgMatches::update_from_arg_matches_mut

fn has_subcommand(__clap_name: &str) -> bool

Test whether Self can parse a specific subcommand