Research
The evidence behind the design, gathered on 14 and 15 October 2026. Each note marks what it could not verify.
| Note | What it covers |
|---|---|
| Amp UX and Orbs | Amp’s public docs, design history and Orbs, and what users say. Source of the UX patterns, from public material. |
| Amp’s TUI, observed | The real Amp CLI watched in a terminal, cell by cell, plus its official images: layout, palette in hex, glyphs, spacing, keys, empty and error states. The reference for the launcher. |
| Herdr | The closest existing tool: architecture, fidelity limits, and what to borrow |
| Host runtime | zmx, libghostty-vt, the holder design, status detection, resume commands, the manifest |
| Connectivity and providers | System SSH, multiplexing, bootstrap, provider recipes, reconnect, security |
| Client stack | Language and TUI framework, attach hand-off, input filter, clipboard, distribution |
| Developer needs | What developers want from remote and background agent sessions: 156 sources (132 from developers), 27 needs ranked by count and GitHub reactions, sentiment per tool, setup pains and paywalled features (15 October 2026) |
| Competitors | 28 hosted, sandbox and open tools in one feature matrix: prices, paid features, persistence, pass-through, and the paid features we can give for free (15 October 2026) |
| Sandboxes and hosts | E2B first, then Daytona, Modal, Fly, Morph, Coder, Codespaces, Hetzner, GCE, EC2 and local VMs: access, persistence, limits, prices, APIs and fit as a host (15 October 2026) |
| Go project structure | Thirteen well-run Go repos: code layout, tests, CI, releases, docs sites and public repo setup; our target layouts and a migration plan (15 October 2026) |
A clean-room study of the Amp CLI binary, and the screens and images captured for Amp’s TUI, observed, are kept outside git (private/, ignored). They must never enter the history of this repo. Only what they show, described in our own words, appears in the notes and in the UX spec.