- What exactly is Semibot?
- A desktop app on your computer—not a web chat box. Conversations, files, knowledge, and run history live in a local SQLite database. It can research the web, write reports and slides, and open your real code repo to edit files and run commands. Office work and coding share one workbench, so you never shuttle context between two AIs.
- Which operating systems are supported?
- Installers for macOS (Apple Silicon) and Windows x64. The Windows build is currently unsigned, so Windows may warn during install—choose “Run anyway”. Linux desktop builds are still on the release plan. The client auto-updates.
- How is it different from ChatGPT, Claude Desktop, or Cursor?
- They are chat-first or IDE-first. Semibot puts research/writing and repo-level coding in one desktop workbench, plus three things they do not have: a secretary that keeps following up after one instruction; a knowledge library that grows into source-linked Wiki articles; and coding with multi-folder grants, file checkpoints, and unified approvals. Built for people who both produce work and run projects.
- Is my data uploaded to a server?
- Not by default. Conversations, knowledge, memory, and run records are written to local SQLite; secrets go to the system keychain, never plaintext config. Data only leaves when you sign in, call a cloud model, or use a connector. Diagnostics and logs are scrubbed.
- Can it change my code? Could it break my project?
- Yes, and behind gates. You grant working and read/write folders with the system file picker; path escapes (`..`, symlinks, UNC) are rejected. Files Semibot itself edits get checkpoints you can safely undo—later edits by you or an IDE are not blindly overwritten. Every task delivers a ChangeSet: files touched, real diffs, commands and tests run. Git is optional source control; git write commands go through the same approval path, and approving git never grants folder access.
- Is it free?
- The client is free to download. First launch creates a free account with trial quota so you can ask cloud models before configuring anything; you can also bring your own model provider. Connectors that need third-party accounts (mail, Feishu, etc.) use your own authorization.
- Do I need to configure a model myself?
- Not to start. Trial quota is ready on first use. Later you can add providers or your own API keys in Settings. The composer has a “use default model” shortcut, and drafts autosave across session switches and restarts.
- How does the knowledge library work?
- Three layers in one workspace: Browse—real folder tree with file preview; Search—hybrid semantic search across this or all workspaces; Wiki—continuously organized topic articles with traceable sources you can keep asking about. It is not a one-shot frozen report; you watch knowledge take shape.
- What can it do for office work?
- Built-in browser for research, filtering material, writing plans and reports, producing documents, slides, and images. Artifacts land in a productions list with preview and syntax highlighting. Long chats get a question-navigation bar; you can also queue new instructions while it is still answering.
- What can the coding agent actually do?
- Open a real project: scan the codebase in the background (no UI freeze on large repos), read project requirements, plan the change, edit files, run commands and tests. One task can span several authorized folders, not just a single repo. You get a ChangeSet and execution results—not a paste-me code snippet.
- What is the native code graph?
- Semibot’s own TypeScript code-knowledge-graph engine—no Python sidecar, no external MCP. It wires calls, references, and structure together so it reads fewer irrelevant files, ramps faster on unfamiliar repos, and can follow blast radius when reviewing a change. The graph is for navigation; high-risk conclusions still go back to source or reproducible commands.
- What do I see in code review?
- An evidence pack: files changed, real diffs, commands run, test results. Files edited through Semibot’s controlled tools with complete checkpoint evidence support safe undo; full command-level rollback can pair with Git worktrees or isolated runs. You review the evidence, then merge, continue, or roll back.
- What does the secretary do?
- In one line: brief it once, it keeps following up—bringing useful results back and only interrupting when it truly needs you. “Weekly client-A briefing every Friday” or “watch the token-reseller scene and send me your read”—no cron, no required first run. Scheduled checks, project digests, material diffs, and coding change briefs collect on the secretary page; it pings you only for missing input or approvals.
- How is a specialist different from a normal chat?
- Normal chat re-explains what you want every time. A specialist remembers how the job is done. Build different specialists for research, writing, transcription, or organizing; before sending, it checks required skills and can enable them in one click without dropping the message or attachments. You only say what to finish this time.
- What is in the toolbox? Does it connect to Feishu or DingTalk?
- Two kinds: skills (teach Semibot new ways of working) and connectors (reach external services). Messaging covers Feishu/Lark, DingTalk, Discord, Telegram, Slack plus Gmail, Calendar, local folders, and BlueBubbles/iMessage. MCP servers can plug in as tool connectors without inheriting your workspace or system credentials. Everything is managed on the connectors page.
- Can I allow only one folder? Who decides high-risk actions?
- Yes. Authorize specific folders with the system picker; grants are per-directory. High-risk capabilities (send messages, change outside data, write files) require approval by default. You can switch session approval modes—“auto-execute” is clearly marked dangerous and still cannot bypass browser hard gates, system permissions, or folder grants.
- Who is it for?
- People who must produce and run projects: solo developers reading unfamiliar repos and shipping fixes, business or research roles that need scheduled briefs and material tracking, and teams that want one shared assistant with confirmable, auditable steps. Not for casual small talk—Semibot is for getting work done.
