I bounced between Claude Code, Codex, Cursor, Hermes, and DeepSeek enough that “where did I already solve this?” became a weekly tax. Agent Sessions does what it says on the tin: it lets me see those conversations in one Mac app—scroll the history, star favorites, export, and search inside a session—without shipping my transcripts to someone else’s cloud.
Repo: jazzyalex/agent-sessions (open source, local-first).
What it is
Agent Sessions is a local-first macOS app for browsing, searching, and (where supported) resuming coding-agent session history. It indexes what’s already on your machine from Codex, Claude Code, Cursor Agent, Hermes, DeepSeek Harness, Copilot CLI, OpenCode, Pi, and a long list of other formats. Your history stays on the Mac. No telemetry on the transcripts.
Install from their releases DMG or:
brew install --cask jazzyalex/agent-sessions/agent-sessionsWhy I care
Harness sprawl is real. Each tool keeps its own logs in its own shape. When I’m comparing setups—or just trying to recover a prompt that worked last Tuesday—I don’t want sixteen Finder windows and a prayer.
This app treats that mess as a product problem. The README and source read like someone who actually lives in these formats: fixtures, stewards, format gates, and honest “unverified” labels where a resume path hasn’t been proven end-to-end. That’s the kind of thoughtful repo I trust more than a flashy landing page.
What I use it for
- One timeline across harnesses. Claude, Codex, Cursor, Hermes, DeepSeek in my case—same library, same scroll.
- Search that goes past titles. Prompts, responses, tool calls, command output, errors, file paths, and supported image references.
- Favorites and export. Keep the sessions that matter; pull them out when you need them elsewhere.
- Search within a conversation. Not just “find the session,” but “find the moment inside it.”
- Resume when the source supports it. Copy a resume command or open a supported CLI session in Terminal, iTerm2, or Warp—capabilities differ by agent, and the app is explicit about that.
- Quota / cost awareness for Codex and Claude. Per-session burn against 5-hour and weekly windows, plus Session Info (tokens, model, thinking effort, API-equivalent estimates). Useful context; not a substitute for your bill.
What’s thoughtful in the repo
A few signals that this isn’t a weekend wrapper:
- Clear supported sources table (browse/search vs resume), including DeepSeek Harness as a read-only history integration with checksum-validated formats.
- Local indexing—session folders are read, not rewritten.
- Optional network only for things like Sparkle updates and a public model-price list—not for uploading chats.
- Contribution path for missing agents (fixtures without writing Swift), stewards, and Session-Bench style format checks.
If you live in multiple agents, that discipline matters more than another “AI for X” pitch.
Who it’s for
- People running more than one coding harness and losing threads between them.
- Anyone who already cares about local history (same instinct as local chat UIs).
- Operators who want a calmer way to audit “what did we actually ask the agent?” without pasting secrets into a third-party dashboard.
If you only ever use one in-editor agent and never look back, you may not need it. If your week looks like mine, you probably do.
Caveats (straight)
- macOS only.
- Resume and quota features are source-dependent; DeepSeek Harness browsing is not an official partnership, and some resume paths are marked unverified in upstream docs.
- API-equivalent dollar views are estimates, not your subscription invoice.
- I’m describing what the project claims and how it fits my workflow—not an endorsement deal.
Link it to the rest of the stack
If you’re still choosing a default harness, start with how to pick an AI coding harness and the longer harness comparison . Agent Sessions doesn’t replace those tools. It makes living with several of them sane.
Related local-stack reading: Msty and four free local AI chat tools .