CLI: integrations, automations, and evals

These commands connect Memory Layer to humans, agents, built-in loop automations, MCP clients, and evaluation suites.

memory tui

Open the main terminal interface.

memory tui
memory tui --project memory

The TUI gives humans a dense view of memories, query evidence, replacement proposals, agents, watchers, activities, embeddings, errors, and project status.

memory agent

Record which local agents are working on which branches or worktrees.

memory agent start --project memory --task "Graph UI follow-up"
memory agent status --project memory
memory agent status --project memory --include-finished --json
memory agent finish --project memory --summary "Pushed graph UI follow-up"
memory agent finish --project memory --abandoned --summary "Superseded by main"

start reads local Git metadata and records the repository root, worktree path, branch, optional task label, head/base commits, dirty files, agent CLI, session id, writer id, and service endpoint. status warns about same-branch work, same-worktree work, dirty-file overlap, dirty active workspaces, and stale heartbeats. finish marks one workspace completed or abandoned and records the final head, dirty files, pushed state, optional summary, and optional merge commit.

memory watcher

Watchers attach to active agent sessions and record useful runtime state.

memory watcher manager enable --dry-run
memory watcher manager status
memory watcher run --project memory
memory watcher status --project memory
memory status --project memory

The watcher manager is the preferred local automation path for Codex-linked sessions. Legacy per-project watcher services and foreground watcher runs remain compatibility paths.

memory loops

Operate the built-in loop automation control plane.

memory loops list
memory loops show memory_hygiene --project memory
memory loops enable memory_hygiene --project memory --mode suggest_only --explicit-user-approval
memory loops run memory_hygiene --project memory --dry-run --reason "inspect cleanup candidates"
memory loops runs --project memory
memory loops inspect <run-id>
memory loops approvals --project memory
memory loops memory-proposals --project memory --status pending

Use memory loops for the Automations tab workflows from a terminal: definitions, settings, trigger routing, manual runs, context packs, approvals, and loop-produced memory proposals. See Automations for the built-in loop catalog and safety model.

memory automation

Operate the older local automation queue controls where they are still used.

memory automation status --project memory
memory automation flush --project memory --dry-run

Prefer memory loops for the current built-in automation control plane.

memory mcp

Run the local stdio MCP server or inspect service-mounted HTTP MCP.

memory mcp run --project memory
memory mcp status --project memory
memory mcp status --project memory --json

Stdio mode is the normal local Codex/Claude path and must write only MCP JSON-RPC messages to stdout. HTTP MCP is local-first and should require token-bearing clients.

memory eval

Run repeatable memory-quality suites.

memory eval doctor --suite evals/examples/memory-smoke
memory eval run --suite evals/suites/memory-improvement-v1 --condition full-memory --dry-run
memory eval compare --baseline target/memory-evals/no-memory.json --candidate target/memory-evals/full-memory.json --text
memory eval gate --comparison target/memory-evals/comparison.json --policy evals/gates/research-v1.toml

Use dry-runs before shell-enabled suites. Evaluation artifacts are evidence for bounded claims about the suite, model, commit, configuration, retrieval settings, token cost, and latency.