I built scritty because I got tired of agents making me repeat myself. Deep into a problem on Claude, I'd hit the usage limit mid thought or debugging, so then I would go over to Copilot which is now a new empty session, so I'd bring it up to speed by pasting context back in. Hit that wall and then I do the same thing on Codex. Run that one dry and I'm off to Antigravity. Burn that out and don't want to wait for limit refreshes so switch to running local Ollama and repeat. Same project the whole way, and every switch turns into the same tedious, token-wasting chore: copy+pasting old conversation, dropping in screenshots, jamming as much as I can into the prompt and hoping for the best. So I built a terminal they all run inside. It sits where every one of them already is, captures the conversation as it happens, and turns it into one searchable memory that carries across every switch. You run any AI CLI inside scritty. It detects what agent is running from the process itself, tags every exchange with the provider, and indexes it locally into an embedded vector store (swappable if you already run qdrant, pgvector, chroma, or weaviate). Search is hybrid and fully offline. Then it exposes that knowledge base over an MCP server and a CLI, so your agents can query their own and each other's past turns, and you can query it as you work. Two things I use every day: - prompt.toml: Because I own the terminal, I can write my rules once and scritty injects them into every message before it reaches whichever agent is running, plus that vendor's own rule file. Markdown files with rules are great, but their relevance is prone to decay, whereas this is in the agent's face every turn (or toggle it off when you don't need/want it). - Phone sync: same terminal session is live on my phone as a PWA. I start something at my desk, walk away, and pick it up on my phone, both ends in sync. My favorite part. It is local-first and paid. Personal is $19.99/mo and runs entirely on your machine with no cloud account; there is a free 14-day pilot and no permanent free tier. Your agents already keep memory, but it is locked in each vendor's box and metered when you reach back into it; scritty's lives on your machine and is free to query, offline. One agent gives you a searchable memory of your own work; every agent you add shares the same corpus. It scales to teams as a shared enforcement layer and a federated, access-controlled, auditable knowledge base, with a per-org control plane for members, billing, seats, audit, license, SSO and SAML, which is a part I really care about coming from regulated environments (banking, healthcare, public sector). Sessions are private by default and you opt into what gets shared. And because capture is at the terminal and not inside any vendor, the agent vendor is swappable: Claude this quarter, Codex or Antigravity the next, nothing lost and no data migration, because the knowledge base is yours and agents rotate through it. There's a 3:09 demo on the site. The whole point is I never start cold again: whoever I'm working with (agent or dev team) can pull up everything the last one already figured out. Happy to answer anything.

Fin Startups get Fin free for a year + 93% off Intercom

Promoted

scritty

Maker

📌

8h ago

how does it actually capture the conversation from agents like Codex and Copilot, do you have to wrap the calls or does it hook into the terminal session itself?

1h ago

The context loss when you switch from Claude Code to another agent mid problem is exactly what kills me, so pulling it all into one searchable memory over MCP is a great idea. Does it keep the full transcript searchable or summarize once a session gets big? Congrats on shipping.

15m ago

scritty

Shared, searchable memory for every AI coding agent

Shared, searchable memory for every AI coding agent