Releases

This page announces each new release of Enkaidu with a skimmable tour of what’s new. For the complete and granular record, see the release notes on GitHub.

0.9.14 Latest

This release makes network and web access more precisely yours to control: network access inside the cordon is now granted per command, the Web tools gain host policies, and the markdown fetch tool parses HTML properly and tells you when it truncates.

Core capabilities

Per-command network access in the cordon - shell_command gains network_allowed_commands: a list of (prefix-matched) commands that run with network access when the cordon is enabled. Requires cordon.policy.allow_network: true; every other allowed command runs without network. Allow ops broadly but grant network only to ops test, for example. Details · Configuration
Host policies for the Web tools - http_get_web_page_as_markdown and http_get_web_page each support allowed_hosts (allow all if unset), denied_hosts, and allow_private_hosts (default false) to control which servers they can be used to reach. Web toolset

Improvements

Real HTML → Markdown conversion - http_get_web_page_as_markdown now parses incoming HTML properly when converting to Markdown; content over 32K characters is truncated, and the tool returns the URL plus a truncated: true | false property alongside the markdown. Details

Experimental !

Experimental features may change a lot or even go away entirely.

Sandboxed config loading - when running with --sandbox, Enkaidu no longer loads config/profile from inside the sandbox folder (the folder the model works in); it looks in the parent folder instead, so the config stays under your control. This anticipates re-using past sandboxes, sort of like sessions, in a future release. See Command Line.

0.9.13

This release widens how you can reach your models, makes the files Enkaidu touches one click away, and tightens the keyboard workflow - plus a hardened Web UI.

Core capabilities

azure_foundry LLM provider - connect to Microsoft AI Foundry with a new azure_foundry provider type, targeting Open AI / chat-completion models. Set the AZURE_FOUNDRY_API_KEY and the bare endpoint URI in AZURE_FOUNDRY_ENDPOINT. Configure Azure AI Foundry.
temperature model setting - tune how “creative” a model is per model via settings: temperature (a number > 0 and <= 2, or default). Learn more.
Profile overrides for session: and cordon: - a .enkaidu/config.yml profile can now carry session: and cordon: sections, so a home enkaidu.yml holds your base setup while a project’s profile overrides just what that project needs. Details.
Cordon networking - a new cordon.policy.allow_network flag lets commands run inside the cordon reach the network. It is all-or-nothing at this time, so enable it with care. Details.

Web UI

One-click file viewing - when Enkaidu writes or edits a file, a button labeled with the file’s name appears above the response; click it to open a side panel with the file’s current contents. See it.
Preview rendering - Markdown code blocks and rendered content (diagrams, charts) preview even while streaming, with a source view available. Rich output.

CLI & keyboard

Prompt history navigation - while editing a prompt, use Up/Down to walk back and forward through your history. It is shared, and kept in sync, between the CLI and the Web UI. More.
Command Line reference - a complete, most-common-first reference of the command-line options. Read it.

Security

Hardened Web UI rendering - generated HTML / CSS and SVG diagrams are now sanitized, closing potential vulnerabilities when rendering model-generated content.

Experimental !

Experimental features may change a lot or even go away entirely.

--sandbox runtime mode - run the current session in an isolated, randomly named sandbox folder under $HOME/.local/state/enkaidu/sandbox. See Command Line.

Older releases

For the full history of changes prior to 0.9.14, see the release notes on GitHub.