Obelisk 🤖 Agent Open source

What it does

Obelisk indexes your past coding-agent sessions into one local SQLite database and makes them queryable by your agent and browsable by you. It reads transcripts from Claude Code, Codex, DeepSeek Harness, Kimi Code, OMP and Pi, and folds them into a single schema with a source field on every row. It has two sides that share the index: a CLI and an agent skill that lets a coding agent search and query its own history in plain language, and an Electron desktop app for humans to browse sessions, manage saved memories, and view usage stats and weekly recaps. The point is to answer questions like "what files did we change to fix that auth bug, and why" without re-reading whole transcripts.

What is inside

• An indexer covering sessions, messages, tool calls, subagents, workflows and workflow agents, with FTS5 full-text search across all layers
• An /obelisk agent skill: the agent writes a JS query against the index, runs it locally, and answers from the JSON result; a core API of search(), context(), sql() and helpers such as fileHistory, failures and workflows
• A memory layer where an agent can propose a Markdown note that is registered against its source sessions and recalled later
• A desktop app with session browsing, a token-usage heatmap and shareable recap cards

Works with

The agents whose transcripts it reads, most relevantly Claude Code and Codex. The /obelisk skill is invoked inside a Claude Code session. Requires Node.js 22.13+.

How to install

npm install --global @obelisk-apps/cli
obelisk --version

obelisk install

Then run /obelisk <your question> in a session. A curl installer and prebuilt macOS app releases are also available.

Maintenance and safety

AGPL-3.0 licensed and actively developed. It reads and writes a real index under ~/.obelisk, so back it up before testing destructive rebuilds. Everything stays local; the memory layer asks for approval before saving a note. The desktop app currently ships prebuilt only for macOS, though the source runs on Windows and Linux.

Who should use it

Developers who run many agent sessions and want to recover past decisions, track which files churn, or audit what subagents concluded, and anyone who wants their history searchable by the agent itself.