Changelog

Changelog

What changed in each release of LLeMbas Server and LLeMbas CLI. Both start their public history here, at 1.0.0.

LLeMbas Server 1.0.0

LatestFirst public release

The first public release of the server: a self-hosted web UI and API for local and remote language models, with agent chats that run on a linked LLeMbas CLI, a library, schedules, voice and images, and the people-and-permissions model to share it.

Highlights

  • Streaming chats with reasoning, attachments, @ references, / commands, folders and search across everything.
  • OpenAI-compatible and Anthropic connections; every model named provider/model.
  • Agent chats on a linked LLeMbas CLI: Manual, Edit, Auto and Plan modes, approval, question and plan cards, a terminal panel, and one session shared between browser and terminal.
  • A library of knowledge bases, notes, memory and skills, with keyword and semantic search.
  • Web search, your own HTTP tools, MCP servers, helpers, schedules and reports, web push.
  • Image generation through ComfyUI, live HTML prototypes, dictation and read-aloud.
  • Users, groups, permissions, quotas, credits with plans, read-only sharing, data groups and an audit log.
  • The /v1 API with scoped tokens, the library over MCP, and an API-only mode.
  • Installable as an app; nine interface languages; Moria and Shire themes.
  • Docker image and compose file; a systemd + nginx installer with web updates on a stable or edge channel.

Release on GitHub

LLeMbas CLI 1.0.0

LatestFirst public release

The first public release of the terminal agent: a compiled binary for Linux on x64 and arm64, signed releases that update themselves, and a link to a LLeMbas server that lets the same sessions be worked from the browser.

Highlights

  • Five API dialects: OpenAI-compatible chat, OpenAI Responses, Anthropic Messages, Gemini and native Ollama.
  • The TUI: streaming Markdown, folding reasoning, inline diffs, a context meter, talking mid-task, six themes.
  • Manual, edit, plan and auto modes, permission rules per tool, and a hardline of commands refused in every mode.
  • Git-native projects: every prompt snapshotted, /undo, /redo, checkpoints, /diff, /commit, /branch.
  • Plan mode with approval, subagents (optionally in their own worktree), a task board and a decision log per project.
  • Memory, skills as commands, searchable sessions, knowledge bases and notes; MCP servers with OAuth.
  • Voice in and out, personality and custom instructions, /changelog and /release.
  • lembas login to a LLeMbas server: its models, voice, search, library and credits; lembas service for agent chats from the web.
  • Self-update on a stable or beta channel, with signed checksums; lembas uninstall.

Release on GitHub