Memorix

Show CodeGraph Memory provider and index status for the current project.

Inspect, preview, or archive bounded checkpoints recorded around a host-native context compaction. Use only to debug or explicitly review compaction continuity. These are lifecycle records, not durable memories or transcript backups. CLI equivalent: memorix checkpoint list|show|context|archive.

Find and merge similar observations to reduce memory bloat. Uses text similarity to cluster related observations by entity+type, then merges them into single consolidated records. Use action="preview" to see candidates without changing data, action="execute" to merge. Example: 10 similar gotchas about Windows paths → 1 consolidated gotcha with all facts preserved.

Build a prompt-ready working context pack for a coding task. Combines relevant memories, CodeGraph Memory facts, freshness warnings, suggested reads, and verification hints. After a complete memorix_project_context brief, provide purpose only when deliberately expanding beyond it.

Maintain a bounded, append-only continuity ledger for a coding task. Use start for requirements, record for decisions/verification/risks, show or list to inspect, and close for the final outcome. Pass the returned taskId to memorix_project_context so the current brief includes this ledger.

Launch the Memorix Web Dashboard in the browser. In HTTP control-plane mode, this reuses the current control-plane dashboard. In stdio/standalone mode, it starts the standalone dashboard server.

Scan active memories for duplicates, contradictions, and outdated information using LLM analysis. Automatically resolves redundant memories. Requires LLM to be configured (set MEMORIX_LLM_API_KEY or OPENAI_API_KEY environment variable). Without LLM, falls back to basic similarity-based consolidation.

Fetch full observation or mini-skill details by typed ref (~500-1000 tokens each). Always use memorix_search first, then pass refs exactly as shown (for example typedRefs: ["obs:42"]). Legacy numeric ids are supported, but typedRefs are preferred.

Inspect the bounded, project-scoped provenance cards behind Memorix context. Use list for a compact index, get/events for one source, sync to repair the projection, and stale when known source files changed. Cards expose sourceRef, locator, capturedHash, and freshness.

Record or inspect auditable memory outcomes. Positive verification/use strengthens recall; correction, conflict, source change, or weaken lowers it. Use revoke with targetEventId to undo one prior signal without deleting audit history.

Show aggregated metrics from recent Memory Formation Pipeline runs. Reports value scores, resolution actions, fact extraction rates, and processing times.

Build a compact, prompt-ready memory graph context packet for the current project. Use this for broad memory overview, project memory graph, or task-specific memory grounding. It returns high-signal memories, entities, relations, and risks as background context, not instructions.

Create a structured handoff artifact when passing work to another agent. A targeted handoff is visible only to its sender and recipient; a broadcast handoff is team-visible. The handoff is stored as a durable observation (immune to archival) and a notification message is sent to the recipient. Use this when completing a task and another agent should continue, or when you want to leave context for whoever works on this next.

Legacy compatibility path: import caller-provided image bytes into the controlled local asset library, then store a text retrieval projection. For durable media operations, prefer memorix_media or the media CLI.

Manage the reviewable project Knowledge Workspace. Use it only for deliberate knowledge operations: initialize a local or versioned workspace, review source-backed claims, compile proposals, lint, apply a reviewed proposal, or manage canonical workflows. It is intentionally absent from the default micro/lite profiles.

Use the controlled local media library for an explicit import, attachment, inspection, or MiniMax generation request. Assets stay outside the Git worktree and enter normal memory only when attach is explicitly true. Use the CLI for destructive removal, quota cleanup, and direct generation. MCP generation is disabled by default and requires MEMORIX_MCP_MEDIA_GENERATION=1 after the operator reviews provider billing.

Get a full snapshot of your project coordination state in one call. Returns: your agent info, watermark (new observations since last session), inbox (unread messages), tasks (your in-progress, available to claim, completed, failed), and team roster (active agents). Use this to decide what to work on next.

Build a compact Memory Autopilot brief for the current coding task. Schedules Code Memory refresh when needed, includes Start here files, reliable code-bound memories, stale/suspect cautions, and verification hints. Use this at the start of a new coding turn or after switching tasks.

Promote observations to permanent mini-skills that never decay and are auto-injected at session start. Action "promote": convert observation(s) to a mini-skill. Action "list": show all active mini-skills. Action "delete": remove a mini-skill by ID. Mini-skills are project-specific specialized knowledge derived from your actual memories — gotchas, decisions, fixes that generic online skills cannot provide.

Mark observations as resolved (completed/no longer active). Resolved memories are hidden from default search but can still be found with status="all". Use this to mark completed tasks, fixed bugs, or outdated information so they don't pollute future context.

Show memory retention status or archive expired memories. action="report" (default): show active/stale/archive-candidate counts. action="archive": move expired observations to archive file (reversible). action="stale": list stale observations with full retention explanation. Uses exponential decay scoring based on importance, age, and access patterns.

Scan project for agent rule files (Cursor, Claude Code, Codex, Windsurf, Antigravity, Gemini CLI, Copilot, Kiro, OpenCode, Trae), deduplicate, detect conflicts, and optionally generate rules for a target agent format. Without target: returns sync status report. With target: generates converted rule files.

Search project memory. Returns a compact index (~50-100 tokens/result). Use memorix_detail to fetch full content for specific IDs.

Search past reasoning traces to understand WHY decisions were made. Returns reasoning memories that explain the thought process behind technical choices. Use this when revisiting code, questioning a design decision, or looking for precedent on how similar problems were solved before.

Get context from previous coding sessions. Use this after compaction to recover lost context, or to manually review session history. Returns previous session summaries and key observations.

End a coding session with a structured summary. This summary will be injected into the next session so the next agent can resume work seamlessly. Recommended summary format: ## Goal [What we were working on] ## Discoveries - [Technical findings, gotchas, learnings] ## Accomplished - [OK] [Completed tasks] - [PENDING] [Pending for next session] ## Relevant Files - path/to/file — [what changed]

Start a new coding session. Returns a compact continuation card with the latest handoff and a few memory references. Call this at the beginning of a session to track activity; retrieve a referenced memory only when it is relevant. Any previous active session for this project will be auto-closed. By default this is lightweight: it binds the project, opens a session, and avoids dumping full history into the new context. Coordination identity is opt-in via `joinTeam: true` or a separate `team_manage` join call. IMPORTANT for HTTP/control-plane mode: pass `projectRoot` with the absolute path to y…

Memory-driven project skills. Action "list": show all available skills from all agents. Action "generate": auto-generate project-specific skills from observation patterns (gotchas, decisions, how-it-works). Action "inject": return a specific skill's full content for direct use. Generated skills follow the SKILL.md standard and can be synced across Cursor, Windsurf, Claude Code, Codex, Copilot, Kiro, Antigravity, OpenCode, and Trae. OpenClaw, Hermes Agent, and Oh-my-Pi receive skills through their official bundle/plugin/package setup lanes rather than generic workspace skill copy.

Inspect memcode's native Memorix runtime: canonical project identity, shared aliases, memory counts, embedding/vector status, search mode, retention posture, and native hooks.

Store a new observation/memory. Automatically indexed for search. Use type to classify: gotcha (critical pitfall), decision (architecture choice), problem-solution (bug fix), how-it-works (explanation), what-changed (change), discovery (insight), why-it-exists (rationale), trade-off (compromise).

Store a reasoning trace — WHY you chose this approach, what alternatives you considered, and what outcome you expect. This creates a searchable record of your decision-making process. Use this when making non-trivial technical decisions, choosing between approaches, or solving complex problems. Unlike regular memories that record WHAT happened, reasoning memories record HOW you thought about it.

Suggest a stable topic_key for memory upserts. Use this before memorix_store when you want evolving topics (like architecture decisions, config docs) to update a single observation over time instead of creating duplicates. Returns a key like "architecture/auth-model" or "bug/timeout-in-api-gateway".

Deep retrieval: expand chronological context around a specific observation — distinguishes explicit memory evolution, hook activity traces, and git-backed facts. Natural follow-up after session L1 routing hints (hook traces) or L3 evidence pointers (git memory).

Export or import project memories. Action "export": export observations visible to the current agent and project sessions (JSON or Markdown). Action "import": import from a JSON export (re-assigns IDs, skips duplicate topicKeys).

Migrate your entire workspace environment between AI coding agents (Cursor, Windsurf, Claude Code, Codex, Copilot, Gemini CLI, OpenClaw, Hermes Agent, Oh-my-Pi, Kiro, Antigravity, OpenCode, Trae). Syncs MCP server configs, workflows, rules, and skills across IDEs. Action "scan": detect all workspace configs. Action "migrate": generate configs for target agent (preview only). Action "apply": migrate AND write configs to disk with backup/rollback.

Open specific nodes in the knowledge graph by their names

Description is built at runtime.

Read the entire knowledge graph

Search the project memory knowledge base for relevant decisions, bugs, architecture notes, gotchas, or other engineering context. Use this when the user asks about project history, design rationale, known issues, or technical details.

Search for nodes in the knowledge graph based on a query

Store a new memory observation to the project knowledge base. Use this when the user wants to save a decision, gotcha, bug fix, architecture note, or any engineering context worth remembering.

Advisory file locks to prevent conflicting edits. Auto-releases after 10 min TTL. Action "lock": acquire lock. Action "unlock": release lock. Action "status": check lock status.

Register, unregister, or list agents in the project coordination state. Action "join": register this agent (returns agent ID and instance ID for reactivation). Action "leave": mark agent inactive, release locks. Action "status": list all agents with roles and capabilities, plus role occupancy. Action "listRoles": show defined roles for this project. Action "addRole": define a new role for this project. Action "removeRole": remove a role definition.

Send, broadcast, or read messages between agents. Durable: messages survive restarts and reach inactive recipients. Action "send": direct message to one agent. Action "broadcast": message all agents. Action "inbox": read this agent's inbox.

Create, claim, complete, or list tasks in the team task board. Supports dependencies. Action "create": create a task. Action "claim": assign to yourself (atomic, race-safe). Action "complete": mark done with result. Action "list": show tasks.

Update an existing memory observation by its ID. Use this when the user wants to modify or add to an existing memory. Provide the ID and the fields to update.

Write content to a file. Creates the file if it doesn't exist, overwrites if it does. Automatically creates parent directories.