scritty Review (2026): Pricing, Features & Honest Verdict

scritty is a terminal emulator that captures every CLI coding agent you run into one searchable memory. Verdict: the only tool solving cross-agent memory where the agents actually live. Best for: solo developers running Claude Code and Codex side by side. Price: from free, Personal $99.99 once. Rating: 7.5/10

What is scritty?

scritty is a terminal that sits underneath whatever CLI agent you launch and records the conversation as it happens. Run Claude Code in one tab and Codex in another, and both sessions land in the same store, each tagged with the provider that produced it. Press Ctrl+Shift+M and you search across all of it without leaving the terminal. The pitch is narrow and correct: your Claude session about an OAuth bug and your Codex session two weeks later about the same bug should be one query, not two archaeology projects.

We found it on the 2 July 2026 Product Hunt board, and its public repository was created on 19 June 2026, so this is a young product with a short track record. What makes it interesting is where it intercepts. Most attempts at agent memory need the agent to cooperate, through a plugin, a wrapper or an SDK you build against yourself. scritty reads the process running in the terminal rather than sniffing the prompt text, which means it captures agents that have no idea it exists. Launch Aider, Ollama, Copilot CLI or Grok inside it and the capture happens anyway.

What Are scritty’s Key Features?

Agent-agnostic capture at the terminal

The terminal is the one boundary every CLI agent shares, and scritty claims it. Because detection happens at the process level, what you submitted and what that pane printed afterwards are recorded together and attributed correctly. No vendor cooperation is required and no vendor can break it with a release. It also runs inside tmux, resolving the owning client and recording agents in separate panes as separate conversations, which is the detail that tells you someone actually used this in anger.

Hybrid search across every agent and session

Search is vector plus keyword, fused with Reciprocal Rank Fusion, running on dual offline ONNX embeddings: one tuned for code, one for prose. Offline matters here, because it means search works with no API key and no outbound call. The vector backend is pluggable, with a built-in embedded engine by default and qdrant, pgvector, chroma and weaviate behind a single trait, with dimension and fingerprint enforcement and atomic blue/green revision swaps. That is more care than most products at this price take over their index.

One corpus, reachable three ways

The store is not a side panel you read. A human at a keyboard, an agent over MCP, and a shell script all read and write the same data with the same behaviour. The MCP server runs over stdio for local agents or Streamable HTTP for remote and multi-machine ones, with bearer-token auth and per-tenant routing. The practical result is the demo scritty leads with: one agent answering from another agent’s prior session. That is the feature nothing else on this list does.

Standing rules that apply to every agent

You write your instructions once in a prompt.toml file and they get injected on every send, rather than politely suggested once at the start of a session. It also reads each vendor’s native rule file, so .claude/CLAUDE.md, AGENTS.md, copilot-instructions.md and GEMINI.md all still work. Every section toggles live from the terminal, so you can flip a constraint off for one message and back on for the next. If you have ever re-pasted the same five rules into three different tools in one afternoon, this is the part that pays for itself.

The same live session on your phone

One process serves two live views. Scan the startup QR code and your desktop session opens on your phone, with scrollback, memory panel and search intact. It is not a read-only mirror: keystrokes from the phone reach the same PTY. Switch a tab on the desktop and the phone follows. Security is tight by default, with TLS always on and no plain-HTTP mode, bearer-token gating that stores only the token hash, a CIDR allowlist, and an explicit opt-in before anything leaves your LAN over a Tailscale Funnel.

How Much Does scritty Cost?

scritty starts free and the paid entry point for one developer is $99.99 paid once, with no subscription attached. The free download has no card, no account and no expiry, and lets you evaluate up to 25,000 exchange-pairs. Personal buys unlimited capture forever plus the first year of updates, after which new builds cost $39.99 a year and are optional. Let that lapse and nothing happens to the copy you own, you simply stop receiving newer versions.

Teams are metered by captured exchange volume rather than by seat. Team Pilot is a free trial capped at 25,000 exchange-pairs or 35 days, whichever arrives first, and needs three or more seats. Above that sit Pro at $249.99 a month for 100,000 federated pairs, Pro Plus at $749.99 for 400,000, and Enterprise from $2,500 for a band described as 1 million and up. A self-hosted option exists and is quoted per organisation rather than listed.

Against the named competition, the solo tier is aggressive and the team tiers are not. Warp charges $20 a month for Build, $200 for Max and $50 per user per month for Business, so scritty’s flat $249.99 Pro undercuts Warp Business at five seats and gets cheaper with every developer you add. Mem0, which sells the memory layer without the terminal, runs a free tier at 10,000 memories, a paid entry at $19 a month and a Pro plan at $249 a month. A one-off $99.99 against those recurring numbers is the strongest argument scritty makes.

What Does scritty’s Per-Pair Math Say About the Team Tiers?

Divide each team price by its included volume and the ladder stops being tidy. Pro works out at $0.0025 per federated pair. Pro Plus works out at $0.001875, which is 25 percent cheaper and exactly what you want to see as you scale. Enterprise then returns to $0.0025 at its published floor, meaning the top tier costs a third more per pair than the tier below it.

The comparison that follows is uncomfortable. Three Pro Plus subscriptions cost $2,249.97 and carry 1.2 million federated pairs, which beats Enterprise’s $2,500 for a band starting at 1 million on both price and volume. Enterprise is not really buying capacity, and scritty should say so. What it actually buys is multi-org federation, dedicated isolation, BYOK, a custom DPA, a 99.9 percent SLA and a hybrid mode that keeps captured data in your own cloud. Those are governance goods, and a regulated buyer needs them. Presenting them inside a capacity ladder makes the step look like a volume discount that runs backwards.

One smaller wrinkle: the page promises “no per-seat pricing, ever” and the ladder honours that, but Team Pilot still requires three or more seats to start. A two-person team has to jump straight to $249.99 a month or stay on Personal licences.

How Local Is scritty’s Local-First Claim?

On Personal, completely, and the commitment is specific enough to check. The licence never contacts the vendor: no heartbeat, no account, no expiry and no kill switch. Captures stay on your machine. That is a stronger promise than most local-first marketing survives contact with, and it is the reason the one-off price makes sense rather than reading as a gimmick.

Above Personal it is partly true. The metered unit on every paid team tier is the “federated pair”, so federation is the product you are buying and data does move. Enterprise offers a hybrid mode that keeps captured data in your cloud, which confirms the default does not. None of that is hidden, but a team reading “local-first architecture” on the pricing header should understand it describes the solo product most accurately.

One thing to be clear about: scritty is not open source. The code lives at a public GitHub repository, which is easy to misread, but GitHub reports no licence file at all, and the site never claims otherwise. A public repo with no licence grants you no rights by default. The repo also carries 11 stars, which is a fair signal of how early this is. Do not buy scritty expecting a community to maintain it if the author stops.

Who is scritty Best For?

Use scritty if you run more than one CLI coding agent and have felt the specific pain of losing a decision made in another tool. It is built for the developer who has Claude Code, Codex and Aider all installed, who works across several repositories, and who wants one query to reach all of it. The free tier is enough to prove the value, and $99.99 once is a low bar for a tool you use every day.

Skip scritty if you use exactly one agent, because your vendor’s own session history already covers you for nothing. Skip it if you need a supported product with a maintenance guarantee, because a two-month-old repository with no licence and 11 stars does not offer one. And think carefully before committing at the team tiers: at $249.99 a month you are buying federation and SSO from a very young company, and the Enterprise floor of $2,500 asks for a level of trust the track record cannot yet support.

Best scritty Alternatives

Warp is the obvious comparison and the honest one. It is a mature agentic terminal with its own AI, free to start, $20 a month for Build and $50 per user per month for Business. It captures its own agent well and other agents running inside it not at all, which is precisely the gap scritty targets. Pick Warp if you want a polished terminal with an agent included; pick scritty if you already chose your agents.

Mem0 sells the memory layer as a service, free up to 10,000 memories, $19 a month to start and $249 a month for Pro with 500,000 add requests. It is more mature and more portable, but it is a library rather than a product: you build the agent that uses it. There is no terminal and no automatic capture, so you write the integration scritty gives you for free.

Your vendors’ own session logs cost nothing and are the real incumbent. Claude Code memory, Codex sessions and Copilot history all exist, and if you live inside one tool they are adequate. They are siloed by vendor, stored one format each, and offer no cross-agent search or embeddings. That is the whole case for paying anything at all here.

Final Verdict: Is scritty Worth It?

For a solo developer running several CLI agents, yes, and the decision is not close. The free tier costs nothing to test, the capture happens without any agent cooperating, the search is better engineered than the price suggests, and $99.99 once with a licence that never phones home is a rare shape in 2026. The MCP server turning the corpus into something your agents can query, rather than just something you can read, is the feature that justifies the product existing.

The team tiers are a different conversation and we would wait. Pro at $249.99 a month is defensible against Warp Business above five seats, but the per-pair math runs backwards at Enterprise, the $2,500 floor buys governance dressed as capacity, and the company behind it is two months old with an unlicensed repository and 11 stars. Buy the Personal licence today. Revisit the team ladder when there is a track record to weigh against it.