SODAX SDKDigital asset flows you can build and deploy with AI.
Promoted
scritty
Shared, searchable memory for every AI coding agent
143 followers
Shared, searchable memory for every AI coding agent
143 followers
scritty is a terminal emulator that captures every CLI agent's conversation (Claude, Codex, Copilot, Antigravity, Ollama), indexes it into one searchable corpus you control, and serves it back to your agents over MCP and to you over the CLI. One session across desktop, browser, and mobile. Your captures stay on your machine.
Works on Windows/Mac/Linux
Interactive
Free Options
Launch Team / Built With
finally a way to stop losing track of what claude told me in that one terminal two weeks ago, the local index feels snappy and the mcp handoff back to my agents actually worked on the first try
scritty
@aslhan345499 exactly the loop we were chasing! thanks for saying it worked first try. local index so it's fast and yours, MCP handoff so your agents can actually read it back.
How does the indexing actually work under the hood when the agents stream output live, do you buffer the full transcript and re-embed after each turn or keep an incremental vector index in sync as the conversation grows?
scritty
@sabanl51707 neither, really. split you drew is right question but answer is third option. granularity is exchange, not whole conversation. we don't buffer full transcript, and we don't re-embed session on each turn. only thing that ever gets re-indexed is the reply currently streaming. it's kept in sync in place as it grows, then it's done. everything already captured stays put. so cost scales with the size of current turn, not length of conversation that way a 200+ turn session isn't any more expensive to keep in sync than a 2-turn one. chunking is content shape aware (code, prose, and stack traces split differently), and unchanged turns are skipped on any rebuild.
scritty
@mdgld it is the coordination layer you're describing exactly as you named it. it's not a new agent and there's nothing to wire in. claude/codex/copilot stay exactly as they are same cli, no wrapper, no plugin, no sdk.
scritty is the terminal they already run in and coordination happens at the pty boundary. so "wire a layer into my agents" is already what's happening. don't want the window? same corpus is an mcp server (scritty serve) + cli and just point your agents at it.
actual value is continuity. stateless agents rederive same repo map every run and decay occurs as a session advances. that's the tax a big repo charges you. scritty holds the map so the next run doesn't pay for it again, and cause capture sits at terminal instead of inside one agent, map built is there when agent wakes up.
on price: for single devs $19.99/mo, heartbeat enforced. basically youre buying the cross session and agent coordination those tools lack. now that everyone bills by the token it pays for itself: inject relevant slice, not wall of prose, and stop paying Nx times for same archaeological dig every task. free pilot is 5k turns / 14 days.
That's pretty interesting! I wonder if this would help w/ distillation of the model?
scritty
@monolithdread interesting angle. scritty isn't a distillation tool itself but capture + memory...that said it does produces the thing distillation needs: a local corpus of raw turns, tagged by which model produced them, that you own and can export. so it is a clean source for building distillation/fine-tune/eval sets from your actual work so that kind of downstream use case is possible. just not the product's intended job today (data model doesn't fight you on it though)
the fact that everything stays local and still feeds back to your agents via MCP is a really thoughtful bit of craft. local-first capture that actually closes the loop with the tools feels rare.
scritty
@nesrin318706 thanks! "closes the loop" is exactly the framing I was going for. local capture + MCP readback is the difference between logs and actual memory. means a lot that it landed. :)
finally something that pulls all my agent chats into one place without sending data anywhere, the MCP search back into claude worked surprisingly well on my session history
scritty
@serhat273347 exactly intended loop :) capture stays local, and same corpus is an mcp server so any agent (claude included) can search its own history back. glad the recall landed and please shout if you uncover useful feedback or desired features.
Hello, this solves a problem I run into quite often. I keep repeating the same project context every time I change AI tools. Having one shared memory across them all feels like a much cleaner workflow
scritty
@mathew_chang exactly. repeating “let me restate the whole project to the next tool” loop was the thing I wanted (maybe even obsessively needed) to kill. shared memory across tools feels much closer to how people/teams actually work than keeping each agent in its own silo.