Head to head

ArgosBrain vs Zed Agent Panel

Zed persists threads. It doesn't extract memory.

What Zed Agent Panel does

Persists full conversation threads in local SQLite. No automatic retrieval, no fact extraction.

What ArgosBrain does differently

Automatic ingest + semantic + graph retrieval, with staleness handling Zed threads lack.

What Zed Agent Panel is

The baseline, stated fairly.

Zed (Zed Industries) persists full threads — conversation + tool-call history — in a local SQLite ThreadsDatabase (parent_id, summary, zstd-compressed JSON data).

There is no automatic retrieval or fact extraction. Users bring context back manually via @-mentions (files, directories, symbols, past threads, rules, diagnostics), the "New from Summary" feature, or MCP servers.

How it actually works

Technical facts.

StorageLocal SQLite only.
RetrievalUser-driven, no automation.
CostZed is free; agent usage is BYOK or Zed hosted. Persistence itself is free.
ScopePer-workspace, local.
Limitationsbuilder.io (2026): "Zed does not currently present skills or memory in a native productized way… you have to compose it via MCP." Issue #38067 loses new agent threads on switch.

Sources: Zed agent panel docs

Verdict

Where each one wins.

↑ Where ArgosBrain wins

Automatic retrieval. Zed requires the user to remember what to @-mention.
Semantic + graph retrieval. Zed is literal thread replay.
Staleness handling. Zed stores threads raw; they rot silently.
The obvious fit. Zed's own docs point to MCP for memory — ArgosBrain is that MCP server.

↑ Where Zed Agent Panel wins

Speed and local-first philosophy align with ours.
Thread history as an archive is a legitimate use case we don't replicate.

When to choose which

Honest recommendation.

Choose Zed Agent Panel if

You want an editor with local-only threads
You prefer manual context curation

Choose ArgosBrain if

You want Zed for the editor + ArgosBrain as the MCP memory server filling the gap Zed's docs admit