continue-transcripts

Convert continue.dev session files to readable, self-contained HTML transcripts.

Features

• Converts continue.dev session JSON files into styled, self-contained HTML pages
• Renders assistant messages as Markdown (code blocks, tables, task lists, etc.) with syntax highlighting
• ANSI → HTML — terminal output with color codes is rendered as styled HTML (One Dark palette)
• Human-readable tool calls — $ command for Bash, key: value for file paths, structured display for multi-arg tools (no raw JSON dumps)
• Tool calls nested under assistant messages — tool calls and their results appear indented under the assistant message that invoked them
• Collapsible thinking & tool results — thinking blocks and tool results start collapsed; expand on click
• System prompt at top — the first system message is extracted and displayed in a collapsible section above the transcript
• Detailed tools reference panel — all tools used in the session are listed with full descriptions, parameters, and usage notes (parsed from the system prompt), each in its own collapsible section
• Token usage tracking — per-message token counts (prompt + completion) with running cumulative totals and session-wide summary in the header (when usage data is available in the session)
• Displays context items (attached files/code) in collapsible sections
• Generates an index.html linking all processed sessions when given a directory
• Optional title-based filtering with --filter
• Parallel processing with configurable worker count (--workers)
• Terminal summary — file counts, elapsed time, and per-file status indicators
• Date+time-prefixed filenames — output files are named YYYY-MM-DD_HHMM_Title.html for easy chronological sorting (falls back to YYYY-MM-DD_Title.html when time is unavailable)
• Filename deduplication — output filenames are truncated to 60 characters with automatic _1, _2 suffixes on collision
• Responsive design — works on desktop and mobile

Quick start

# Build from source (requires Rust toolchain)
cargo build --release

# Convert a single session file
./target/release/continue-transcripts session.json

# Convert all sessions in a directory
./target/release/continue-transcripts ~/.continue/sessions/ -o ./transcripts

# Filter by title and use 4 workers
./target/release/continue-transcripts ~/.continue/sessions/ --filter "auth" -w 4 -o ./transcripts

Output looks like:

Using 4 worker thread(s)
Discovered 12 JSON file(s)
  ✅ transcripts/2025-06-15_1430_Fix_failing_tests_in_auth_module.html
  ✅ transcripts/2025-06-14_0915_Refactor_auth_module.html
  ...
  ✅ transcripts/index.html

═══════════════════════════════════════════
  Summary
═══════════════════════════════════════════
  Files discovered:  12
  Sessions written:  10 (+ 1 index)
  Skipped:           2
  Workers:           4
  Elapsed:           1.23s
───────────────────────────────────────────
  ✅ All files processed successfully!

Installation

Pre-built wheels are available from GitHub Releases. No Rust toolchain is required — just Python >= 3.9 and uv.

With uv tool install (recommended)

Install globally so the continue-transcripts command is always available:

uv tool install continue-transcripts \
  --no-index \
  --find-links https://github.com/curtisalexander/continue-transcripts/releases/expanded_assets/v0.14.0

The continue-transcripts command is then available on your PATH.

To upgrade later (update the version in the URL):

uv tool install --upgrade continue-transcripts \
  --no-index \
  --find-links https://github.com/curtisalexander/continue-transcripts/releases/expanded_assets/v0.14.0

To uninstall:

uv tool uninstall continue-transcripts

With uvx (one-off runs)

Run without installing:

uvx \
  --no-index \
  --from continue-transcripts \
  --find-links https://github.com/curtisalexander/continue-transcripts/releases/expanded_assets/v0.14.0 \
  continue-transcripts ./sessions

This is useful for a quick one-off conversion without permanently installing the tool.

With uv pip install

Install into a specific virtual environment:

uv venv .venv
source .venv/bin/activate
uv pip install continue-transcripts \
  --no-index \
  --find-links https://github.com/curtisalexander/continue-transcripts/releases/expanded_assets/v0.14.0

Building from source

If you prefer to build from source (or need a platform not covered by the pre-built wheels), you can install directly from the Git repository. This requires a Rust toolchain in addition to Python >= 3.9:

uv tool install git+https://github.com/curtisalexander/continue-transcripts

Or build with Cargo directly:

cargo build --release

Pre-built wheel platforms

Wheels are built in CI for the following targets:

Platform	Target
Linux x86_64	x86_64-unknown-linux-gnu
Linux aarch64	aarch64-unknown-linux-gnu
macOS Apple Silicon	aarch64-apple-darwin
Windows x86_64	x86_64-pc-windows-msvc

Usage

continue-transcripts [OPTIONS] <INPUT>

Arguments

Argument	Description
INPUT	Path to a single session JSON file or a directory containing session files

Options

Option	Description
-o, --output <DIR>	Output directory for generated HTML (default: output)
--filter <STRING>	Only process sessions whose title contains this string (case-insensitive)
-w, --workers <N>	Number of parallel worker threads (defaults to number of CPU cores)
-h, --help	Print help

Examples

Convert a single session file:

continue-transcripts session.json

Convert all sessions in a directory:

continue-transcripts ~/.continue/sessions/ -o ./transcripts

Filter sessions by title:

continue-transcripts ~/.continue/sessions/ --filter "auth" -o ./transcripts

Use 8 worker threads for a large directory:

continue-transcripts ~/.continue/sessions/ -w 8 -o ./transcripts

The tool writes one HTML file per session to the output directory, plus an index.html that links to each transcript.

Where are continue.dev sessions stored?

Continue stores data in ~/.continue/ on Linux and macOS (%USERPROFILE%\.continue\ on Windows). Session history files are JSON files within this directory. The exact subdirectory may vary by version — check your ~/.continue/ tree or use find ~/.continue -name '*.json' to locate them.

How it works

1. Discovers session JSON files from the input path (single file or recursive directory scan)
2. Reads and parses files in parallel using rayon (configurable thread count)
3. For each session:
   • Extracts the system prompt (first system message) to a collapsible top section
   • Collects full tool descriptions from the system prompt and renders each tool as a collapsible reference panel with parameters and usage details
   • Tracks token usage per message and calculates running cumulative totals
   • Groups assistant messages with their tool calls and tool results
   • Converts ANSI escape codes in tool output to styled HTML
   • Renders Markdown to HTML with syntax highlighting via pulldown-cmark + syntect
4. Produces a self-contained HTML file per session (all CSS and JavaScript are embedded — no external dependencies)
5. Deduplicates output filenames (truncated to 60 chars, with _1, _2 suffixes on collision)
6. Generates an index.html listing all sessions when processing multiple files
7. Prints a summary with file counts, timing, and status indicators

Inspiration

Inspired by Simon Willison's claude-code-transcripts.

License

MIT