github binaricat/Netcatty v1.1.45
on GitHub

latest releases: v1.1.79, v1.1.78, v1.1.77...
one month ago

Download based on your OS:

OS Download
Windows Setup x64
macOS DMG Apple Silicon DMG Intel X64
Linux AppImage x64 DebPackage x64 RpmPackage x64 ArchPackage x64
AppImage arm64 DebPackage arm64 RpmPackage arm64 ArchPackage arm64

Release Highlights

πŸ€– A much stronger AI foundation

This release is the largest AI infrastructure update Netcatty has shipped so far. Catty, external agents, the Netcatty MCP server, CLI tooling, and future global agents now share a single capability catalog instead of relying on separate tool definitions that can drift over time. That gives each tool a clearer contract for what it can read, what it can change, when it needs approval, and how results should be returned.

The AI turn flow has also been rebuilt around a shared runtime. Turn lifecycle, stopping, tracing, context preparation, tool output storage, and duplicate-result handling now live in one place, so AI sessions should feel more predictable and will be easier to extend safely in future releases.

🧰 Expanded agent tools

Catty now has access to 41 structured tools, covering terminal sessions, SFTP, Vault hosts and notes, snippets, port forwarding, attachments, workspace/session inspection, terminal context reading, URL fetching, web search when configured, and reading stored full output after a large result is shortened. The future global-agent tool surface now has 35 shared tools prepared as well.

Vault automation is more intentional now. SSH host creation and import go through dedicated Vault host tools, while Vault Notes and Host Details notes stay separate. This reduces the chance that an agent puts host inventory in the wrong place, and makes write actions easier to review and approve.

🧠 Context compression designed for long-running work

A lot of effort went into making long AI sessions stay usable instead of failing once the conversation gets too large. Netcatty now estimates context usage with provider-aware budgets, protects the most recent messages, compresses older structured content, and has a dedicated recovery path for oversized requests. When compaction happens, the active permission mode, current session scope, user goal, and pending tool-output handles are re-injected so the agent can continue from the same working state instead of losing the thread.

Large tool outputs are handled with the same principle: the model gets a useful preview first, while the full result is kept behind a handle that can be read later. This keeps terminal dumps, SFTP reads, Vault note bodies, and other bulky results from overwhelming the next AI step, without throwing away information the agent may still need.

Catty can now read a bounded slice of the active terminal screen or scrollback, which makes it much better at understanding what just happened in a terminal without dumping an entire session into the chat. Files attached to the current message are also registered before the turn starts, so attachment tools can reliably see dragged-in files.

πŸͺͺ Cleaner permission and approval behavior

Auto/Confirm permission modes are now handled consistently across settings, sync, bridges, MCP status, and agent prompts. Confirm-mode permission memory has a clearer UI and simpler copy: users can review saved allow rules, add rules manually, or approve matching future operations more quickly.

🧩 Richer AI result cards

Tool results from Catty, external SDK agents, MCP, and Netcatty CLI paths now feed into the same artifact presentation layer. Vault host and note results can render as useful cards, AI note cards open directly in the side panel, missing-note cases remain understandable, and terminal context reads get compact terminal result cards instead of noisy raw output.

πŸ–₯️ Terminal and connection improvements

The terminal also got several practical upgrades: Windows cold-start font rendering is fixed, Telnet now handles NVT newlines and echo mode more correctly, multiline compose/snippet sends no longer drop content after the first line, terminal clears respect the scrollback preservation setting, manual session logging is available from the toolbar, and SSH network stats now include latency.

Port forwarding through jump hosts now carries known-host verification context correctly, hostname whitespace is handled more safely, Ctrl+Tab skips pinned-tab shortcuts properly, and dynamic sidebar/tab titles refresh more reliably.

πŸ›‘οΈ Important reliability fixes

This release includes fixes for several high-impact issues: empty DeepSeek tool-call fields, hidden Catty attachments, external tool artifact cards, side-panel note navigation, and macOS local terminal startup when the node-pty helper loses executable permission. Together, these fixes make both AI-driven workflows and day-to-day terminal startup more robust.

What's Changed

New Contributors

Full Changelog: v1.1.44...v1.1.45

Check out latest releases or
releases around binaricat/Netcatty v1.1.45

Don't miss a new Netcatty release

NewReleases is sending notifications on new releases.

Get notifications