Agent tools

The agent works through tools. Full registry (16 tools):

Files & commands (local)

ToolWhat it does
list_filesBrowse a folder inside the workspace
search_filesMulti-threaded recursive search by file name and content (the agent uses it instead of manual traversal)
read_fileRead a file (with offsets for large files)
create_or_update_fileCreate or fully rewrite a file
patch_fileSurgical edit (find/replace a fragment)
delete_fileDelete a file
run_commandRun a shell command in the workspace
web_searchWeb search: works without keys (DuckDuckGo); a Tavily key in Settings improves stability and quality

Plan (Plan mode)

ToolWhat it does
propose_planPropose a plan for approval. You edit/approve the plan via a card, and only then the agent starts working

AI memory

ToolWhat it does
save_memorySave a fact to long-term memory (visible in all chats)
save_memory_projectSave a fact to project memory (bound to the workspace folder)
delete_memoryDelete a memory entry by ID prefix

Entries are injected into the context on every request and consume no FDAgentAi tokens. The limit depends on the plan (Free 100 / Lite 150 / PRO 250 entries).

Servers (SSH / SFTP / FTP)

ToolWhat it does
ssh_execRun a command on a connected server
remote_list_filesBrowse files on a server (SFTP or FTP)
remote_upload_fileUpload a workspace file to the server
remote_download_fileDownload a file from the server into the workspace

Connections are managed in the «Servers» section; secrets are encrypted. Any server operation always requires approval — even in Full Access mode.

Parallel execution

Independent operations (file reads, search_files, web_search, remote_list_files, memory tools) run in parallel — noticeably speeding up long runs. Mutating operations (writes, deletes, commands, plan) are strictly sequential.

Approvals

File writes/deletes, run_command and all remote operations show an approval card with a before/after diff (for file edits). In Plan mode the plan is approved first, then the dangerous steps.