Comparing OpenCode Export and Claude Copy | Generated by AI

Home 2026.03

Question: Is the analysis comparing OpenCode’s /export (session transcript) feature versus Claude Code’s /copy button accurate — specifically that OpenCode exports the full agentic “journey” while Claude Code only copies the last response?

Answer:

Your analysis is mostly correct, but a few details need refinement based on the actual documented behavior of both tools.

OpenCode’s /export Command

OpenCode’s TUI provides an /export command that exports the current conversation to Markdown and opens it in your default editor (using the $EDITOR environment variable).

The export dialog (DialogExportOptions) allows the user to set the output filename (defaulting to session-<id-prefix>.md), toggle inclusion of thinking blocks, tool details, and assistant metadata, and choose to open the transcript in $EDITOR without writing to disk.

So it is indeed a full-session, structured export — not just the last turn. The granular toggles (thinking blocks, tool details, metadata) confirm it captures the entire agentic process, not just final responses.

Claude Code’s /copy Command

Claude Code’s behavior is more nuanced than simply “copy the last response”:

Claude Code’s /copy command copies Claude’s most recent response to your clipboard. If the response has code blocks, you get an interactive picker to grab individual blocks instead of the full response. You can also pass a number to go further back — /copy 2 grabs the second-to-last response.

However, Claude Code does have a full export option too:

The /export command in Claude Code writes the full conversation as plain text directly to disk — no prompt, no dialog. The export is plain text only, no JSON, no structured format. It is a transcript.

The key gap is that Claude Code has no prominent built-in export button in its UI; conversations are stored in ~/.claude/projects/ as JSONL files. This has led developers to build third-party tools to fill the gap.

The Real Distinction

Your core point holds up well:

Feature OpenCode /export Claude Code /copy
Scope Full session, all turns Last response (or N responses)
Format Structured Markdown with toggles Plain text or clipboard
Tool details Toggleable (thinking blocks, tool calls) Not included by default
Thinking traces Toggleable Hidden by default; thinking traces are not included when manually copying terminal sessions
Design philosophy First-class TUI export dialog Per-response clipboard helper

Transcripts from Claude Code sessions capture extremely important context — what was asked for, what Claude suggested, decisions made, and Claude’s own justification for the decisions it made while implementing a feature. The community clearly values this, which is exactly why tools like claude-code-transcripts (by Simon Willison) were created to compensate for Claude Code’s lack of a polished built-in export flow.

One Correction to Your Framing

OpenCode is not built on Zig or “OpenTUI/Zig” as described. OpenCode’s interactive TUI is built with Bubble Tea (a Go framework) for a smooth terminal experience, and it supports multiple AI providers including Anthropic Claude, OpenAI, Google Gemini, AWS Bedrock, Groq, Azure OpenAI, and OpenRouter.

So: your functional comparison is accurate, but the architectural description (Zig/OpenTUI) is incorrect — OpenCode is a Go + Bubble Tea project.

References:

Back Donate