Waku 0.1.0 bets the product is the control plane, not another coding agent

Waku's current version, 0.1.0, is a native desktop control plane for the coding-agent command-line tools developers already run, built in Rust on GPUI and collecting sessions, transcripts, tool activity and repository checkpoints inside one local application ^[1][2]. The bet behind it is that developers using several agents do not need another agent, model or editor, but a reliable place to see what each one did, interrupt it, continue the conversation and recover when a task goes wrong ^[3].

That decision shows up in the plumbing. Waku avoids Electron, keeps its working data on the user's machine, and connects to existing agent CLIs through their structured interfaces, translating those different event streams into one provider-neutral model ^[4]. According to its product page, the application connects through interfaces including stream-JSON, JSON-RPC and live events, choosing the strongest available connection for each provider ^[5]. The screenshot Waku published appears to show Amp, Claude Code, Codex, Cursor, OpenCode, Grok and Pi in the driver layer ^[6], seven providers whose sessions are meant to land on a common timeline while keeping their own authentication and execution paths ^[7][18].

The consequence of that architecture is a maintenance obligation rather than a modelling one. Each supported CLI can change its event schema, session format and resume behavior independently, so Waku's value depends on how well its adapters absorb those differences and how quickly they recover when providers ship updates ^[8]. The distribution detail is therefore load-bearing: Waku says the application is signed and notarized, with Sparkle handling automatic updates through binary deltas ^[16]. A control plane that cannot patch itself quickly is a control plane that is wrong about the state of your repository.

The feature that separates this from a dashboard sits under the transcript. Every prompt creates a checkpoint of the working tree under a hidden Git reference, and a developer can restore the repository and the associated provider conversation together ^[10]. That pairing is the point. Agents read files, run commands and edit code across turns, so reopening an old message without restoring the corresponding tree leaves conversation and working state out of sync ^[11]. Treating the prompt as a transaction boundary is a stronger claim about how agent work fails than most agent UIs make.

The interface follows the same supervision model: Command-N starts a session, Return queues another prompt while an agent is working, Command-Return steers the active turn and Escape stops it ^[9].

On the native side, GPUI is the GPU-accelerated interface framework created by the team behind Zed ^[12]. Waku's homepage says the app offers instant startup and smooth navigation through long transcripts, though it publishes no independent performance measurements ^[13]. Waku also says projects, transcripts, sessions and provider identifiers stay on disk, with no Waku account, telemetry or hosted intermediary ^[14]. Worth being precise about what that does not cover: the underlying agents are not local, and prompts, code and tool output remain subject to whichever agent CLI and model provider the user selects ^[15].

What to watch: how fast adapters are repaired after a provider changes its event schema or resume behavior ^[8], and whether checkpoint restore stays trustworthy as the number of concurrent sessions grows. The interface layer around coding agents is becoming a product category of its own ^[17], and in that category reliability of the recovery path, not the provider count, is the differentiator.