Auto-replied with thread context · 1.2sThree steps. One thread.
Set it up once. Forget it. The memory builds itself in the background, and shows up when you need an answer.
What clicks immediately.
The reaction is usually the same: this does not feel like another assistant. It feels like the work already remembers enough to keep moving.
The Slack reply made it obvious. I did not need to reopen Linear, Docs, and three old threads just to remember what changed.
The MCP part clicked when Claude Code could start from actual work state instead of a fresh prompt recap.
Most AI products look clever in the demo. This felt different because the thread behind the work was already there.
I like that it starts personal. It does not ask the whole team to change behavior before it becomes useful.
Weekly updates stopped feeling like reconstruction work. The raw material was already sitting across tickets, notes, and conversations.
The strongest part is not the answer. It is that the answer comes with the thread behind it, so you can trust it.
A few things people ask first.
Before they trust a memory layer with the thread behind their work.
Not really. Peppermint is the memory layer underneath the model. It turns the thread behind your work into usable memory — facts, summaries, routines — so Slack, Claude Code, Codex, Linear, or the next tool can work from real context instead of a blank box.
A read-only window into the parts of your memory you allow. Other tools (Claude Code, Codex, your own scripts) can query the same context your assistant uses, without copying it out of your machine.
Source context — the raw threads, tickets, and docs — stays local by default. When you choose to share, only the summary leaves your memory layer, and it goes to the surface you pick.
Models without memory restart every conversation. Peppermint gives them the same continuity your team already has — what was decided, what shipped, what is blocking — so the model starts where the work actually is.
Memory does not switch with you because it never lived inside one tool. The same context shows up in Slack, in Notion, in Linear, in your editor.
No. The memory layer runs locally with a small footprint. Indexing happens in the background, ranked by recency, with a hard cap on CPU and disk.