Devlog

Devlog copies all your coding dingus transcripts to a Third Location so you have access to them after their default retention period deletes them.

Supports Claude Code, Opencode, and pi.

Install

npm install -g @cianfrani/devlog
devlog init

devlog init asks where to store your archive (default: ~/devlog) and which agents to configure the MCP server for.

Commands

devlog                  # copy sessions, then index them for search
devlog archive          # only copy sessions to your archive
devlog index            # only index archived sessions into SQLite
devlog mcp              # start the MCP server (stdio)
devlog init             # first-time setup

Each stage skips sessions that haven't changed since the last run. Use archive or index when you only need one stage.

Options

--verbose    show per-project and per-session details
--debug      extra noisy diagnostics
--rebuild    with devlog or index, wipe the database and re-index everything

MCP server

After running devlog init, you can search your session history from inside your coding agent.

Tools: search, list_sessions, get_session, schema, query

If you don't want to use an MCP, write a skill that describes where the sqlite or sessions are stored.

search past sessions for "auth middleware"
list sessions for project "quoting-ui"
get session <session_id>

Raw SQL works too:

SELECT project, COUNT(*) as n FROM sessions GROUP BY project ORDER BY n DESC

How it works

Archive reads each agent's local session data and writes it to ~/devlog/ as normalized JSONL. Claude Code and pi store sessions as files, so devlog copies those directly. Opencode stores sessions in a SQLite database now, so devlog extracts and converts them.

Index parses the archived JSONL into a SQLite database with full-text search (FTS5, porter stemming). Before content is written to SQLite, devlog does a best-effort scrub of common credential patterns plus exact matches for sensitive environment variable values such as OPENAI_API_KEY and GITHUB_TOKEN. The MCP server reads from that database.

Where redaction applies

Redaction runs only on content written to the SQLite index, which is the surface your coding agent queries through the MCP server. Archived JSONL files in ~/devlog/ are copied byte-for-byte from the source session files — they are never rewritten or scrubbed. This is deliberate:

• The archive is meant to be a faithful backup of what the agent actually saw. Rewriting it would make it impossible to diff against the source or recover the original context.
• The index is the queryable surface. Scrubbing there is where it prevents secrets from leaking into search results or MCP responses.

If you need the archive itself to be clean, scrub it out-of-band (e.g., wipe a session file, then devlog index --rebuild). The scrubber is intentionally narrow: it helps avoid casually surfacing secrets, but it is not a full PII/privacy sanitizer.

Session sources

pi extension records

pi writes generic extension records (custom and custom_message) alongside its conversational messages. Devlog indexes these from their public envelope only — it never interprets any third-party extension's private payload schema. A custom entry's data is serialized as opaque JSON inside a <pi:custom> block, so nested values stay searchable but are displayed verbatim rather than reformatted. Raw archived JSONL files are never rewritten by indexing; upgrading across an index-version bump rebuilds the local SQLite cache on the next devlog index, making previously dropped extension state searchable again.

File layout

Follows the XDG Base Directory spec:

Rebuild the index from the archive at any time with devlog index --rebuild.

Configuration

~/.config/devlog/config.json (created by devlog init):

{
  "archiveDir": "/Users/you/devlog",
  "dbPath": "/Users/you/devlog/index.db",
  "excludeProjects": ["fidelio", "scratch"]
}

excludeProjects examples:

• "fidelio" matches -Users-you-src-tries-2026-02-17-fidelio
• "scratch" matches /Users/you/Code/scratchpad

Requirements

• Bun — the CLI runs on the Bun runtime.

Contributing

Working from a checkout: bun install && bun run build:cli, then invoke ./dist/cli.js directly.