Nanobot

Nanobot drafts a project workbench while exposing unfinished UI tests

Nanobot's September 8 interface drafts separate project work from topic history, add live settings and bring usage charts into the terminal. The project-navigation draft still reports ten failing layout tests, and the expanded Memory settings page has an unresolved compatibility gap.

← Back to homeOriginal source ↗

Nanobot's latest interface proposals make its small agent core look increasingly like a workbench. A September 8 project-navigation draft separates the sidebar from an ever-growing topic history, while adjacent settings and terminal-usage proposals expose more runtime behavior directly to users. The important news is not polish alone. Workspace scope, unsent drafts, memory policy and usage visibility are becoming explicit product surfaces, with unfinished verification disclosed rather than hidden behind screenshots.

The facts

- The navigation draft gives Topics, Projects and Automations fixed destinations, with topic history in the main panel. - Entering a project scopes navigation and Ctrl/Cmd+K search; project identities derive from existing workspace scopes. - Unsent drafts preserve attachments and quoted context separately across topics and projects, but only in the current app session's memory. - The settings proposal exposes 48 configuration fields and separates idle-compaction from long-term-memory controls. - The terminal /usage proposal shows up to eight recent logical model rounds and distinguishes unknown cache usage from zero. - The project-layout test report lists 60 passes and ten failures, while the settings draft discloses an unresolved blank Memory page on a mismatched development backend.

Why it matters

A lightweight runtime does not eliminate the need to organize several ongoing jobs. It shifts the question to whether the interface accurately reflects workspace and memory boundaries. A project picker that silently loses an attachment, or a memory switch that merely hides a panel, would undermine the promised simplicity. These drafts make the state contract more explicit: draft persistence is session-local, disabling long-term memory preserves existing files, and workspace changes are deferred until active work finishes.

Current

Inspected on 2026-09-09. The Nanobot stable-release baseline is v0.3.0 published 2026-07-25T08:08:47Z. The main source was open (not merged) when captured. Release metadata is a version boundary; an open proposal, a merged mainline change and a packaged release remain different states.

Evidence

The primary evidence is HKUDS/nanobot PR #5710 (https://github.com/HKUDS/nanobot/pull/5710). Supporting context comes from HKUDS/nanobot PR #5704 (https://github.com/HKUDS/nanobot/pull/5704); HKUDS/nanobot PR #5705 (https://github.com/HKUDS/nanobot/pull/5705). These pages were inspected directly. Related project records explain the source cluster; they are not independent replications of contributor test claims.

Operator take

Review the three proposals as a workbench direction, not one finished release. Test navigation with an unsent attachment and quote, inspect search scope inside and outside a project, and verify that a settings change takes effect in the runtime without disrupting active work. The terminal usage pane deliberately stays local and should not become an agent message when opened during a turn. That is a useful interaction boundary to preserve. There is no reason to accept failing layout tests merely because adjacent runtime tests are green.

Caveat

All three proposals were open at capture. The navigation head had not passed full validation, and the settings draft explicitly needed a compatibility fallback and development verification. Reported terminal tests used isolated transcripts and a mock gateway, not a live model billing comparison. These are proposals, not shipped workbench guarantees.

All three proposals were open at capture. The navigation head had not passed full validation, and the settings draft explicitly needed a compatibility fallback and development verification. Reported terminal tests used isolated transcripts and a mock gateway, not a live model billing comparison. These are proposals, not shipped workbench guarantees.