> Direct answers to what teams ask before adding memory to an AI agent: what it is, why vector search is not enough, MCP memory servers, rollback and audit.

Canonical page: https://bilinc.space/answers

- Guides

# Straight answers about agent memory.

Short reference pages for the questions that come up before a team commits to a memory layer. Each one answers the question first and explains the mechanism second.

[Install Bilinc](https://bilinc.space/install) [Compare options](https://bilinc.space/compare)

## Start with whichever one you are actually asking.

**What is AI agent memory?**
The state an agent carries between runs — decisions, constraints, preferences, and outcomes — and the operations that keep it correct over time.[Read the answer](https://bilinc.space/answers/ai-agent-memory)

**Why isn’t vector memory enough?**
Similarity search retrieves text that looks related. It cannot tell you who wrote a fact, where it came from, what it replaced, or how to undo it.[Read the answer](https://bilinc.space/answers/why-vector-memory-isnt-enough)

**What is an MCP memory server?**
A Model Context Protocol server that exposes memory operations as tools, so any MCP-capable client can commit and recall state without custom glue code.[Read the answer](https://bilinc.space/answers/mcp-memory-server)

**How do you roll back an agent memory write?**
Snapshot before a risky run, inspect the diff afterward, and roll back the known-bad state instead of hand-cleaning rows.[Read the answer](https://bilinc.space/answers/agent-memory-rollback)

**How do you audit what an agent remembered?**
Keep the operational record attached to the write — which run, tool, or operator produced it, and what it replaced.[Read the answer](https://bilinc.space/answers/audit-agent-memory)

**See it in the client you already use.**
Bilinc connects through MCP, the Python SDK, or the CLI. The integration guides cover Claude Code, Codex, Cursor, opencode, and any MCP-capable runtime.[Coding agents](https://bilinc.space/for/coding-agents) [MCP setup](https://bilinc.space/docs/mcp)

**Or start from the workflow.**
Coding agents, research agents, stateful product agents, shared team memory, and incident recovery each hit the memory problem from a different direction.[Use cases](https://bilinc.space/use-cases) [Pricing](https://bilinc.space/pricing)
