Waku Coding Agent Desktop

Installation is per-user, Linux does not require root, externally managed daemon exposure is loopback-only by default, and browser origins must be configured exactly. This supports reasonable least privilege, but the daemon still owns workspace filesystem, Git, terminal, and attachment operations without evidence of per-capability isolation. The README explains local storage and the desktop/daemon/cache/token boundaries, but the release workflow contains a configurable analytics endpoint without documenting telemetry contents, activation, or opt-out, so data-flow transparency is not complete. A stable authentication token and uploaded attachments are sensitive, yet no encryption-at-rest, file-permission, retention, or deletion policy is shown. Dependencies use version constraints, one patched dependency is commit-pinned, and update artifacts use signing; points are deducted because core GPUI dependencies follow mutable Git branches and no dependency audit or vulnerability scan is evidenced. Filesystem, Git, updater, and remote-daemon effects are described, but confirmation gates for every consequential action are not shown. Git-aware rewind and referenced uninstall guidance provide partial recovery, while attachment, database-migration, and updater rollback details are absent. Repository ownership, a security email, sponsorship, and GPL provenance are stated; the publisher remains unverified by the named registry, so maintenance attribution is not treated as fully established.

The README, Cargo manifest, and release workflow are broadly consistent about versioning, platforms, architecture, and packaging. Protocol checks, tag/version matching, and exact artifact paths reduce release inconsistency, but the supplied evidence lacks enough runtime implementation and tests to corroborate every product claim. Rust, Bun, platform toolchains, supported agent CLIs, and Linux/Windows prerequisites are identified, and release builds pin Rust; mutable Git branches and reliance on external services reduce dependency-availability certainty. Workflows emit explicit errors for missing versions, mismatched tags, absent artifacts, and empty releases, but almost no evidence shows user-facing diagnostics for desktop, daemon, or agent-session failures, so failure messaging receives only partial credit.

The product is clearly aimed at desktop users working with local coding agents and covers projects, independent sessions, follow-up steering, model/access selection, and Git rewind, though guidance is not segmented by role or complex deployment scenario. Capability boundaries are unusually concrete: supported agents, unavailable folder-picker and PTY behavior with an external daemon, macOS-only browser/computer-use features, and native Linux/Windows capabilities are all stated, justifying full credit. Automatic CLI detection, explicit daemon exposure, and exact browser-origin configuration show reasonable trigger control; the deduction is that complete rules for starting agents, executing tools, or approving high-impact operations are not provided. macOS, Linux, Windows, Wayland, X11, MSVC, multiple architectures, and a portable Windows package are explicitly addressed, providing strong environment-fit evidence.

The README is organized into installation, supported agents, highlights, architecture, development, sponsorship, and licensing, with clear links to responsible crates, apps, and supporting documents. Installation paths cover all three desktop platforms; Linux documents rootless placement and references manual installation and removal, while Windows offers installer and portable variants. Waku, waku-daemon, waku-core, waku-client, and waku-protocol names remain consistent across documentation and manifests. The main deduction is the absence of complete end-to-end usage examples, UI examples, and an FAQ. Platform and external-daemon limitations are explicit, but migration, telemetry, recovery, and security limitations are not consolidated. GPL-3.0-only is consistent across README, Cargo metadata, and the complete LICENSE, earning full credit. Cargo versioning, tag validation, release artifacts, and CHANGELOG extraction demonstrate a formal release path, but the actual changelog was not supplied. SECURITY.md identifies a latest-release support policy, private reporting channel, and maintenance email, and RELEASING.md is referenced; however, publisher identity is unknown and no explicit maintenance team or long-term governance structure is shown.

The product centralizes multiple agents' projects, sessions, transcripts, model settings, follow-up messages, diffs, files, and Git rewind in a native interface, making the claimed output useful for ordinary workflows. The deduction is that no screenshots, complete interaction examples, or representative outputs are supplied to establish presentation quality. A shared interface, structured provider protocols, and session continuity offer clear marginal value over operating several CLIs separately, but some value remains assertion-only and the external-daemon mode lacks the folder picker and PTY. No Waku account, local-first storage, per-user installation, and GPL licensing reduce adoption cost; users must still install and authenticate separate agent CLIs, while development requires Rust, Bun, and platform toolchains, and automatic updates plus a possible analytics endpoint introduce operational and privacy tradeoffs.

Architecture claims commonly point to named crates, configuration paths, protocol-generation commands, and release workflows, while version and license statements map directly to manifests and LICENSE. The deduction is that the supplied material omits referenced core implementation, platform guides, CONTRIBUTING, RELEASING, and CHANGELOG files, leaving several feature and security claims only partially traceable. README, Cargo.toml, SECURITY.md, and both workflows strongly corroborate platform, version, signed-update, maintenance-channel, and release-path claims, justifying full cross-source credit. Current, experimental, and unavailable capabilities are usually distinguished clearly; deductions remain because promotional claims such as “fast” are unsupported and the “no remote service required” framing does not explain the analytics endpoint visible in release configuration.