Obelisk

Obelisk indexes the session history of coding agents into one local SQLite database so that an agent can query its own past work and a person can browse it. Sources include Claude Code, Codex, GitHub Copilot, DeepSeek Harness, Hermes Agent, Kimi Code, OMP, Pi, and ZCode. Every provider is projected into the same schema, with sessions, messages, tool calls, subagents, workflows, and memories, and full-text search covers all layers.

The agent side is a CLI plus an agent skill. When you ask a question such as which files changed in the last fix for a bug, or which tool calls failed recently, the agent writes a JavaScript query against the index, runs it locally, reads the JSON result, and answers in plain language. Query helpers cover search, context, raw SQL, sessions, summaries, workflows, failures, and file history. When a retrieval produces a conclusion worth keeping, the agent proposes a markdown memory file, and after you approve it the memory is registered and recalled in later sessions.

The app side is an Electron desktop app for browsing sessions with readable tool calls (diffs, terminal output, file views), managing memories, viewing token usage and activity heatmaps, and generating weekly or monthly recap cards. The index rebuilds incrementally and watches provider transcript locations for live refresh.

Each provider has its own adapter. Branching and compaction state is preserved where the source records it: Pi, OMP, and ZCode mark superseded history as inactive so normal queries omit it, while Claude Code and Codex transcripts carry no such state.