neuron-explorer
Steering-first neuron explorer for Qwen3.5-4B: push every MLP neuron, keep every rollout, browse the results
Review, change, and verify real repositories from Desktop or terminal.
Developer toolsRust · TypeScriptGitHub @JimmyDaddyOnlineOwner unconfirmed
Have you tried it? One tap is enough. No sign-in needed.
A TUI-first coding agent for reviewable, resumable repository work.
Taken from the repository README
Loading the full README. The saved excerpt is shown below.
Reviewable edits. Resumable sessions. Desktop or terminal.
A Rust coding agent with first-class Desktop and TUI experiences.
View macOS prereleases · Watch Desktop + TUI demos · Changelog
Sigil is under active development and continuous iteration, so features may be unstable and change without notice. The website and user docs follow main; packaged releases can lag behind. Check Installation and the Changelog before relying on a newly documented feature.
Desktop and TUI show bounded live previews while an answer is streaming, then replace them with the saved result. ReopRead the full README ↗[!WARNING] Sigil is under active development and continuous iteration, so features may be unstable and change without notice. The website and user docs follow
main; packaged releases can lag behind. Check Installation and the Changelog before relying on a newly documented feature.
| Work in context | Stay in control |
|---|---|
| Desktop or terminalUse the surface that fits your workflow while keeping the same conversation, task, approval, and recovery semantics. | Review before riskInspect approvals and diffs before writes, commands, network access, or external integrations proceed. |
| Resumable sessionsReturn to saved work and recover interrupted tasks without silently rerunning an unfinished tool. | Models and tools, your wayChoose among supported providers, add MCP integrations, and enable repository-aware assistance when you need it. |
| Large outputs stay inspectableSigil keeps bounded conversation views while preserving policy-safe tool output as session-scoped artifacts for precise, paged follow-up reads. | Cache-stable contextHistorical tool output ages deterministically before semantic compaction, so long sessions shed token pressure without rewriting the active cached prefix. |
| Commands have one lifecycleFinite checks run in the foreground with live progress; persistent or interactive work uses explicit terminal tasks and event-driven waits. | Approval matches real effectsOne immutable permission plan binds parsed commands, targets, containment, policy, approval, audit, and execution across Desktop and TUI. |
Desktop and TUI show bounded live previews while an answer is streaming, then replace them with the saved result. Reopening a session restores committed history; a truncated preview is not the complete answer. In Desktop, use “View complete message” to read the saved text page by page. Plan revision can keep accepting input while details load, and an interrupted cleanup remains visible as unfinished recovery work. Only the current session and protocol formats are supported; older formats are rejected without automatic migration.
TUI keeps the rail compact. Press F4 (Alt-I / Shift-F2 also work), or click F4 details in the rail, for scrollable workspace, session, and usage details even in narrow terminals. F3 opens settings while keeping your draft. F1 lists the keys; Ctrl-O expands focused tool output.
In TUI, Alt-B opens completed conversation turns for branching. Choose a turn, then edit your next draft in the new session; branching does not send it or restore workspace files. See the TUI guide.
npm install -g @sigil-ai/sigil@beta
cd /path/to/your/project
sigil
Quick Setup opens when configuration is missing or invalid. No values are reused or migrated from an invalid file: Desktop Settings and the TUI offer an explicit reviewed replacement that is published only while the live file is still invalid. Choose a provider and model, add authentication, and run sigil doctor if anything looks incomplete. The Quickstart takes you from a first read-only task to a small reviewed change.
Prefer a native app? The GitHub prerelease provides signed and Apple-notarized macOS DMGs for Apple Silicon and Intel. See Installation for the exact asset names and update path.
Automatic Task routing is the default: a model chooses Chat, PlanReview, or Task through a typed decision. A plan waits for your approval; a direct Task can begin when the configured provider and executor support it. Release evaluation informs new-install defaults for proactive read-only agents, and does not grant execution permission. File, shell, network, MCP, external-directory, and merge permissions still apply. Use routing_policy = "manual" to disable automatic handoff. See Advanced Configuration.
| Guide | What it covers |
|---|---|
| Visual tour and TUI user guide | Desktop workspaces plus TUI controls, approvals, sessions, and recovery. |
| Configuration | Common setup paths and exact fields. |
| Providers and MCP | Models, authentication, and integrations. |
| Safety, permissions, and privacy | Decisions, limits, and data handling. |
| Troubleshooting | Symptoms, checks, and recovery actions. |
| Reference | Commands, keys, paths, and exit behavior. |
Project status · Contributing · Developer docs · Security · MIT License