scritty
Summary
Five AI coding agents, one project, and zero shared memory — that's the default state of every multi-tool terminal workflow, where what Claude decided about auth last week is invisible to Codex today.
Scritty is a terminal emulator that detects which AI CLI is running inside it, captures every exchange, and builds a single searchable corpus across all agents and sessions — stored on your machine, served back over MCP and CLI. The browser-sync feature runs a token-secured web server inside the same process, so the identical PTY and memory panel open in any browser tab or on a paired phone, byte-for-byte in sync. Teams running four or five agents on one codebase report the biggest gain: no re-pasting decisions between tools. The ceiling appears when you need cross-team memory sharing; the per-tenant control plane is a paid-only feature, and offline-first teams hitting network gaps rely on local Ollama models via the PWA path.
Bottom line: Scritty earns its place the moment you're running three or more AI CLIs on the same codebase and re-explaining context between them; it breaks down as a standalone solution when an engineering org needs centrally audited, multi-tenant memory without standing up the control plane tier.
Community Performance Report Card
No community ratings yet. Be the first to rate this tool!
Pros
Sign in to edit- Automatic agent detection and per-provider tagging across Copilot, Codex CLI, Claude Code, Aider, and Ollama, so you stop manually labeling or re-pasting context when you switch tools mid-session.
- Single searchable corpus shared across all agents and sessions, which means 'what did we decide about auth two weeks ago' becomes a CLI query instead of a scroll through four separate vendor logs.
- Local-first capture with no external data transmission by default, so teams with data-residency requirements or air-gapped workflows can use the tool without negotiating a privacy policy.
- Live browser and phone sync over a token-secured in-process web server, so a session started on desktop continues on a phone without an app install, account creation, or a separate sync service.
- Offline operation via local Ollama and PWA support, so a lost network connection does not halt work — the same memory and terminal remain accessible.
Cons
Sign in to edit- Multi-tenant memory isolation and audited admin overrides — the features an engineering org needs before rolling this out across teams — are behind the paid-only control plane tier. A small team hits this wall the moment a second developer needs their own isolated corpus on shared infrastructure, and the practical next step is either upgrading or running separate scritty instances per developer.
- There is no outbound API for querying captured memory outside the MCP server path. Tooling that needs to pull session history into a CI pipeline, a custom dashboard, or a third-party search index has no documented path to do so, and teams with that requirement are looking at competitors that expose a REST or webhook layer.
- The browser and phone sync depend on the desktop process staying alive. Drop the desktop machine — a laptop that closes, a process that crashes — and the mobile session loses its host. Teams that need persistent, always-on terminal access independent of a local machine end up running this on a server or switching to a cloud-based terminal alternative.
About
- API Available
- No
- Self-Hosted
- Yes
- Last Updated
- 2026-07-03T03:36:27.070Z
Best For
Who it's for
- Developers using several AI CLI coding agents
- Users who need offline access via local models
- Teams wanting audited, tenant-isolated agent memory
- Anyone frustrated by per-vendor context silos
What it does well
- Running multiple CLI agents on one codebase while sharing memory
- Searching past decisions across all agents and sessions from terminal or phone
- Applying consistent rules and instructions to every agent without re-pasting
- Continuing a terminal session from desktop to mobile or browser
Integrations
Add notes, reviews, and benchmarks so the next visitor gets a clearer picture.
Compare scritty
Spotted incorrect or missing data? Join our community of contributors.
Sign Up to ContributeFrequently Asked Questions
- Is scritty free?
- scritty has a permanent free tier alongside paid upgrades. You can keep using a baseline version indefinitely without paying.
- Is scritty open source?
- No — scritty is a closed-source tool. Source code is not publicly available.
- Can I self-host scritty?
- Yes. scritty supports self-hosting on your own infrastructure.
Curated lists that include this category
Scritty is a terminal emulator built around one workflow: run any AI CLI inside it, and every exchange gets captured, tagged by provider, and indexed into a shared searchable corpus. The agent detection is automatic — Copilot, Codex CLI, Claude Code, Aider, Ollama, and others are recognized without configuration. That corpus is then served back two ways: to agents themselves over MCP, and to you directly from the CLI or a built-in search panel. Captures stay local by default; no data leaves your machine unless you wire up the control plane.
The differentiating feature is the live browser and phone sync. Scritty embeds a token-secured web server in the same process as the terminal. Opening the URL — on another tab, another device on your network, or externally over Tailscale Funnel — gives you the exact same PTY, the same scrollback, and the same memory panel. This is not a read-only mirror: input from the browser or phone drives the same session. A pairing QR at startup gets a phone connected in seconds, with no app store install required.
Scritty fits tightest for individual developers or small teams who context-switch between multiple AI coding agents on the same project and are losing work to per-vendor silos. The offline path via local Ollama and the PWA makes it usable without a network connection. Where it strains: teams that need multi-tenant memory isolation and admin audit trails must use the control plane feature, which the vendor describes as a paid-only capability. There is no API surface exposed for programmatic integration outside of the MCP server path the docs describe.