sivtr

sivtr is a local-first shared memory workspace for humans and agents. It turns the work around a project—terminal commands, command output, AI-agent conversations, tool results, and copied context—into searchable, selectable, referenceable memory that both you and your agents can reuse.

Use it beside the terminals and agents you already have. It gives their local work one shared memory workspace.

For the full workflow, install both pieces: the sivtr CLI/TUI and the bundled sivtr-memory skill. The CLI captures and retrieves memory; the skill teaches agents to use that memory before asking you to paste context.

What sivtr is not

sivtr is not:

• a terminal emulator;
• a tmux replacement;
• a hosted transcript service;
• another agent runtime.

What sivtr helps with

• Keep the output, not just the command from pipes, subprocesses, shell integration, and local agent transcripts.
• Browse long logs comfortably in a keyboard-first Vim-style TUI.
• Copy the last useful thing as input, output, bare command, or full block.
• Search local agent conversations from Codex, Claude Code, Hermes, OpenCode, or Pi when you need an old decision or explanation.
• Let agents start from evidence so “fix the terminal error” can begin with the latest captured failure instead of a paste request.
• Jump back to the source behind a summary, search hit, or handoff note.
• Save search results as variables like @last and @failures, then reuse them in follow-up commands.
• Launch memory pickers from the terminal, tmux, VS Code, Windows hotkeys, or generated desktop shortcuts.

First useful commands

# Browse command output as reusable workspace memory.
cargo test 2>&1 | sivtr

# Let sivtr run the command and capture combined output.
sivtr run cargo test

# Copy the latest recorded command output.
sivtr copy out

# Copy the latest assistant answer from an agent provider.
sivtr copy claude out
sivtr copy codex out
sivtr copy opencode out
sivtr copy pi out

# Search current workspace memory.
sivtr search agent --match "panic" --format timeline

Common workflows

Goal	Start here
Install the CLI + skill	Installation
Learn the daily path	Quickstart
Understand the model	Mental Model
Capture output	Capture Terminal Output
Copy recent commands	Copy Command Blocks
Reuse agent memory	Work with AI Sessions
Teach agents the memory workflow	Skills and Reusable Procedures
See practical community workflows	Playbooks
Search and dereference memory	Search and Show Results
Open pickers quickly	Launch Pickers and Hotkeys
Check exact syntax	CLI Reference

Mental model

sivtr has two layers:

Layer	What it describes
What happened	Terminal output, command blocks, agent conversations, tool results, and local history.
How you reuse it	TUI browsing, search, copy, show, diff, skills, playbooks, and memory variables like @last.

Terminal sources produce command blocks. Agent providers produce conversation blocks. Selectors like 1 and 2..4 pick recent items for copy commands. Search results can be saved as variables such as @failures, then shown, expanded, or piped into the next command.

Local by default

sivtr reads local shell logs, local history, and local agent transcripts. Shared Codex trees are opt-in through explicit export and configuration. See Data Locations for where records live.