Quickstart
sivtr has two everyday paths:
- For humans: open the
sivtrTUI panel to browse terminal output and agent sessions together, search across sources, and copy the exact part you need. - For agents: install the bundled
sivtr-memoryskill, then use CLI commands such assivtr search,sivtr copy --print, andsivtr showto retrieve local workspace memory.
1. Install the CLI and skill
Section titled “1. Install the CLI and skill”Follow Installation first. The important point is that agent workflows need both parts:
- the
sivtrCLI must be onPATH; - the bundled
sivtr-memoryskill must be installed withnpx skills add Ariestar/sivtr --skill sivtr-memory -g.
After that, the agent can be told to “use sivtr first” and will have a concrete procedure for searching local memory.
2. Put terminal history into the workspace
Section titled “2. Put terminal history into the workspace”If this is your first time using sivtr, install shell integration once:
sivtr init powershell# or: sivtr init bash# or: sivtr init zsh# or: sivtr init nushellRestart the shell, then work normally:
bun run buildcargo testgit status --shortRecent commands and output will now appear in the workspace. Agent sessions (Claude, Codex, Hermes, OpenCode, Pi) are read from their local session directories.
3. Open the unified workspace TUI
Section titled “3. Open the unified workspace TUI”Run:
sivtrThis is the most important entry point. It is not terminal-only, and it is not tied to one agent provider. It puts local workspace memory into one interface.
The workspace TUI has four panes:
| Pane | Purpose |
|---|---|
| Source | Choose a source, such as terminal, Claude, Codex, Hermes, OpenCode, or Pi. |
| Sessions | Choose a terminal record or agent session. |
| Dialogues | Choose a dialogue turn or command block. |
| Content | Inspect input, output, tool results, or message content. |
Common keys:
| Key | Action |
|---|---|
0 / 1 / 2 / 3 | Focus Source, Sessions, Dialogues, or Content |
j / k | Move inside the focused pane |
h / l | Move to the previous / next pane |
Space | Toggle the current source, session, or dialogue |
/ | Search the workspace |
n / N | Next / previous search result |
i | Copy input or question |
o | Copy output or answer |
y | Copy input + output |
c | Copy bare command when available |
: | Set a line filter for the next copy |
t | Open the current content in full view |
z | Toggle focused pane fullscreen |
? | Help |
q / Esc | Back or quit |
Workspace search supports prefixes:
| Query | Scope |
|---|---|
error | Content |
#build | Dialogue or command-block title |
>release | Session title |
4. Open only one kind of content
Section titled “4. Open only one kind of content”The unified workspace is the default entry point. When you only want one source, use a more specific picker.
Terminal command blocks only:
sivtr copy --pickOne agent provider only:
sivtr copy claude --picksivtr copy codex --picksivtr copy hermes --picksivtr copy opencode --picksivtr copy pi --pickThese are filtered workspace views, not the main entry point.
5. Copy directly when you know what you need
Section titled “5. Copy directly when you know what you need”When you already know what to fetch, you do not need to open the TUI:
sivtr copy # latest command + outputsivtr copy out # latest output onlysivtr copy cmd # latest command onlysivtr copy out 2 # previous command outputsivtr copy 2..4 # recent rangeRead agent sessions:
sivtr copy claude out --printsivtr copy codex out --printsivtr copy hermes out --printsivtr copy opencode out --printsivtr copy pi out --printFor agent use, prefer --print so the command returns text instead of opening an interactive UI or only writing to the clipboard.
6. Let the agent search workspace memory first
Section titled “6. Let the agent search workspace memory first”When something fails, tell the agent:
Fix the terminal error. Use sivtr first.The agent should search the same workspace memory:
sivtr search terminal --match "error|failed|panic|Traceback|Exception|exit code|FAILED" --format json --limit 20sivtr search terminal --status failure --latest 1 --format jsonsivtr copy out 1 --printsivtr copy cmd 1..10 --printThen it can read code, patch the issue, and rerun the smallest relevant verification command.
This is the core value: what you can see in the unified workspace, the agent can also read through commands.
7. Return to exact evidence with refs
Section titled “7. Return to exact evidence with refs”Search results include refs. A ref can point to a terminal command block, an agent session, a dialogue, or a specific content block.
sivtr search terminal --match "build error" --format json --limit 20sivtr show terminal/current/2sivtr show claude/<session>/3sivtr show claude/<session>/3/2Refs let humans and agents return to the same evidence instead of relying on vague summaries.
8. Use the single-output browser when needed
Section titled “8. Use the single-output browser when needed”Besides the workspace TUI, sivtr also has a single-buffer browser for temporary long output:
cargo test 2>&1 | sivtrsivtr run bun run buildThis browser is for one captured output. The bare sivtr workspace TUI is the main entry point for terminal and agent memory together.
Next steps
Section titled “Next steps”- Read the Mental Model to understand sources, sessions, dialogues, blocks, and refs.
- See Playbooks for real workflows.
- Read Browse and Select for full workspace TUI controls.
- Keep CLI Reference open for exact options.