Polylogue

Polylogue is a local archive and workbench for AI work. It captures every session you have with AI tools (Claude Code, Codex, Gemini CLI and other coding agents through their own files and hooks; ChatGPT, claude.ai, AI Studio and other web chats through their exports and a browser extension), keeps the original bytes, parses them into one model of sessions and their lineage, and lets you and your agents search, read, measure and annotate that history from the command line, over MCP or in a web workbench. Agents are first-class users: they get the same powers as you, and every session can start with a brief drawn from the archive.

Polylogue runs alone (standalone, on SQLite) or on Sinex (primary mode), where Sinex stores the bytes and records and Polylogue keeps a local index. Everything you can do is the same in both modes.

Install

With Nix:

nix build github:Sinity/polylogue      # the package, at ./result/bin/polylogue
nix develop                             # a development shell

The flake also provides a Home Manager module (the user service, hooks, MCP registration, the browser extension's host, completions) and a NixOS module; see operations.

With Cargo, from a checkout:

cargo build --release -p polylogue

Polylogue reads each tool's data where the tool writes it, so a running daemon needs no configuration to start collecting. Exports you download (a ChatGPT or claude.ai data export) are brought in with polylogue import.

A tour on the Morrow archive

Morrow is a synthetic week of work on a fictional Rust library, in Claude Code, Codex, ChatGPT and claude.ai, shipped with Polylogue for examples and tests (examples/morrow/). The tour builds a separate archive from it in a new directory and walks through ordinary commands, with each command and its answer written to transcript.md:

polylogue demo tour ./morrow-tour

First commands

polylogue daemon                          # run the daemon (normally a user service)
polylogue                                 # what the archive holds and what is running
polylogue 'repo:morrow since:7d'          # a query: this week's sessions in a repository
polylogue find chrono read                # search for a word and read the matches
polylogue @1 read --to browser            # open the first row of the last list in the workbench

A query comes first and verbs act on its rows; polylogue explain describes the language, and polylogue --help lists commands. Every command exists over MCP and HTTP too.

Where to read next

• docs/: the design, one page per area, starting with architecture.
• docs/reference/: current generated artifacts and planned reference pages. Use polylogue --help and the MCP schemas for commands available in the current build.
• AGENTS.md: how work is done in this repository, its layout and its commands.
• Python client: the declared commands through CLI JSON and the daemon, with unknown values and opaque IDs preserved.

Licence

See LICENSE.

Cloud setup and shell startup, including the explicit native backend, are described in the shared setup guide.