XMEComing Soon

Xanther Memory Engine

Persistent memory for coding agents. XME gives your agent long-term recall — decisions, preferences, and context that persist across sessions, IDEs, and team members. Linked to your codebase architecture via XCE.

Join Waitlist (Discord)See Context Engine

The problem with stateless agents

Every new chat, every new task — your agent starts from zero. It doesn't remember what it tried before, what decisions were made, or what your preferences are. You end up re-explaining the same context over and over. Memory Engine fixes that.

How XME + XCE Work Together

🏗️

XCE — Context Engine

Understands your codebase architecture. Knows what code does and how it connects.

🧠

XME — Memory Engine

Remembers decisions, preferences, and history. Linked to architecture nodes in XCE.

⚡

Together

Agent knows the code AND remembers what happened. Context + memory = true intelligence.

Agent asks about auth module
  → XCE returns architecture context (files, dependencies, patterns)
  → XME returns memory ("last week we decided to migrate from JWT to sessions")
  → Agent has full picture: code structure + historical decisions

Capabilities

Persistent Memory

Your agent remembers past interactions, decisions, and context — even across restarts and IDE switches.

▸Long-term memory store
▸Contextual recall
▸Automatic memory extraction

Cross-Session State

State carries over between sessions. Your agent picks up right where it left off — no re-explaining.

▸Session continuity
▸State snapshots
▸Cross-IDE persistence

Decision History

Track every decision your agent made and why — full audit trail for debugging and knowledge transfer.

▸Decision logging
▸Reasoning traces
▸Queryable history

User Preferences

Learn coding style, naming conventions, and architectural preferences over time. Applied automatically.

▸Style learning
▸Convention enforcement
▸Team-wide standards

MCP Tools

xme_remember

Store a memory node — decisions, preferences, findings, or any context worth preserving.

xme_recall

Query memories by topic, time range, or relevance to the current task.

xme_session_state

Get/set session state — task progress, open questions, next steps.

xme_preferences

Read/write user and team preferences — coding style, conventions, patterns.

xme_history

Query decision history — what was tried, what worked, what was rejected and why.

Use Cases

Multi-day refactors

Agent remembers what it already explored, what approaches failed, and what's left to do — across days and sessions.

Team knowledge sharing

Shared memory means new team members' agents inherit knowledge about the codebase, conventions, and past decisions.

Debugging continuity

Pick up a debugging session days later — the agent recalls the full investigation trail, hypotheses tested, and findings.

Style consistency

Agent learns your team's coding patterns and applies them automatically across PRs — no more style review comments.

Cross-IDE workflow

Start in Cursor, continue in Claude Code, review in Kiro — memory travels with you across every MCP-compatible tool.

Architecture decisions

Record why architectural choices were made. Future agents (and developers) can query the reasoning behind any decision.

Architecture

┌─────────────────┐     MCP/SSE    ┌─────────────────┐
│  Coding Agent   │◄──────────────►│   XME Server    │
│  (any IDE)      │                │                 │
└─────────────────┘                │  Memory Store   │
         │                         │  + Graph Links  │
         │                         │  + TimeSeries   │
         ▼                         └────────┬────────┘
┌─────────────────┐                         │
│   XCE Server    │◄────────────────────────┘
│  (architecture) │   Offline sync: memories
└─────────────────┘   linked to graph nodes

Memory Types:
  • Decisions    — "We chose X because Y"
  • Preferences  — "Team uses camelCase, 2-space indent"
  • Findings     — "Bug was in auth/middleware.py line 42"
  • State        — "Task 3/5 complete, blocked on API review"
  • History      — "Tried approach A (failed), B (worked)"