Search BrigadeStack

Search BrigadeStack pages and docs.

BrigadeStack
Menu

Comparisons

BrigadeStack vs agentmemory: memory or the whole layer?

agentmemory and BrigadeStack both keep memory across coding agents and both do handoffs. Here is where they overlap and where they differ.

Updated November 15, 2026

Short answer: these two overlap. Both keep persistent memory and do handoffs across coding agents, so this is a like-for-like comparison. The difference is scope and how memory gets written. agentmemory is focused on memory that follows your agents: an MCP server with 54 tools, hooks, a local store, and a viewer. BrigadeStack treats memory as one part of a wider operator layer that also keeps your MCP servers and skills in one catalog, files handoffs through a lint and owner policy, and records verify receipts for the work. Pick agentmemory if you want automatic capture and recall, fast. Pick BrigadeStack if you want memory plus config plus work records under a review step, in plain files in your repo.

Where they overlap

Both keep context across sessions and across tools, both work with the common coding agents such as Claude Code and Codex, and both have a notion of handing a task off between agents. agentmemory has a session_handoff prompt. BrigadeStack has writer inboxes and an ingest step. If all you want is “my agent should stop forgetting,” either can do that.

Where they differ

agentmemoryBrigadeStack
Persistent memory across agentsyesyes
Handoffs between sessions and agentsyesyes
MCPships its own server and writes its own entry into agent configssyncs one catalog of any servers into each tool’s config
Skillsships its own 17installs reviewed skills into each harness
Capturehooks capture automatically, with a privacy filter for secretsagents draft handoffs, and a lint checks them
Before a writeno approval step documented. You browse, edit, and forget afterward in the viewersafe, targeted notes file under owner policy. Ambiguous ones wait for review
History and undoaudit trail, version chain, snapshot create and restoregit history of the cards, plus a receipt for each verified run
Storageits own local store (file by default, optional Redis), with an optional MEMORY.md exportplain markdown cards in your repo

BrigadeStack’s bet is that memory alone is half the problem. Your agents also share MCP servers and skills, and all three drift out of sync. BrigadeStack keeps the three together, shows the plan before it writes config, and records a receipt for the work. Safe, targeted notes file on their own, and ambiguous ones wait for a review decision that agentmemory does not ask for. agentmemory also has things BrigadeStack does not: automatic capture of memory from hooks, graph recall, and lightweight action and signal tools for coordinating several agents. BrigadeStack’s optional backend has hybrid search and claim leases, but you run that backend yourself.

When agentmemory fits better

  • You want persistent memory for your agents and not much else.
  • You prefer automatic capture and recall over a review step.
  • You want a viewer for browsing and editing what the agents stored.

When BrigadeStack fits better

  • You want one catalog for memory, MCP servers, and skills instead of three separate setups.
  • You want a plan or review step before config and memory change, and a record of verified work.
  • You want your memory as plain files in git, readable without the tool.

Questions

Can BrigadeStack do handoffs like agentmemory? Yes. Writer agents leave handoff notes in their own inbox, BrigadeStack lints and guards each one, and the safe notes file into one shared memory. See handoff promotion.

Is BrigadeStack heavier to run? It asks you to review writes. You can let safe, targeted notes file themselves and only step in for the ambiguous ones.

Just want to share MCP servers, not memory? See sharing MCP servers across Claude Code and Codex.

Quickstart and GitHub.