Terminal Agent Commands

Terminal Agent provides several commands to help you interact with the terminal efficiently.

Available Commands

Command	Description
`ask`	Query the LLM with a question to get information
`task`	Execute a task using LLM-generated commands
`routine`	Define and run scheduled, unattended agent routines
`daemon`	Run and manage the routine scheduler daemon
`skill`	List, create, and show reusable task skills
`tool`	Manage and execute specific tools
`plugin`	Install and manage plugins
`config`	Configure Terminal Agent settings
`history`	Query your interaction history
`memory`	Store ask memories and inspect task memory

Task Memory

Task runs are indexed into a local task memory database by default. The index is stored at ~/.local/share/terminal-agent/task_memory.db; no task memory data is sent anywhere except as part of the normal model prompts needed to answer your current request.

Task memory is used by agent task in two ways:

Before every model iteration, Terminal Agent searches local task memory with a one second budget and adds up to five compact, relevant memories to the task prompt.
During a task, the agent can use the read-only memory_search tool when new information appears, such as an instruction read from a file or an error message that may have been solved before.

Task indexing is asynchronous and best-effort. Task execution does not fail if memory indexing fails or is delayed.

Task memory stores several record types. The most useful records are tool_success entries because they include the previous tool name, structured tool input, and output. task_completion entries capture the final answer, and task_request entries capture that a similar task was asked even when no command has been learned yet. A completed run indexes exactly one task_completion record — the final answer step is folded into it, including the model's closing thought, rather than indexed separately.

Search ranks actionable memories first. For example, if a previous task for what time is it learned unix {"command":"date '+%I:%M %p %Z'"}, that tool_success memory is preferred over a bare task_request or final-answer-only task_completion. This makes the command visible to the model and to session-log validation. Previous outputs in retrieved memories are historical; for current-state requests such as time, date, files, processes, or git state, the agent should reuse the remembered tool input and run it again rather than copying the old output.

The index uses SQLite FTS5 when available. If the local SQLite build does not provide FTS5, Terminal Agent falls back to local lexical scoring over recent memory records. You can see which path is active with agent memory task stats.

Useful commands:

agent memory task search "print current time"
agent memory task stats
agent memory task clear
agent config set task-memory false
agent config set task-memory true

The session logs under ~/.local/share/terminal-agent/sessions/ remain the authoritative task history. The SQLite task memory database is a searchable index derived from task activity; clearing it only resets what future tasks can recall. The Graphical UI's Task Memory tab browses, searches, and clears the same index.

To confirm a task used automatic memory retrieval, inspect the newest task session log and look for records with "status":"task_memory_retrieval". These records include the retrieval query, match count, and compact matched memories, including tool inputs for actionable matches. Explicit memory_search tool calls also appear as normal tool-call/tool-result records.

Example validation command:

grep -R '"status":"task_memory_retrieval"' ~/.local/share/terminal-agent/sessions

Common Flags

Most commands support these flags:

Flag	Short	Description
`--provider`	`-p`	The LLM provider to use (e.g., "openai", "anthropic")
`--model`	`-m`	The specific model ID to use
`--loglevel`		Set logging level (debug, info, warn, error, etc.)
`--print`	`-x`	Whether to print the response to stdout
`--log`	`-l`	Whether to log interactions to history
`--plain`	`-k`	Display output as plain text (no markdown formatting)

Command Details

For detailed information about each command, see: