Skip to main content
Glama

SessionSift

Free, local-first conversation analysis for Codex, Claude Code and explicitly imported ChatGPT/Claude exports. Browse conversations, inspect tokens and tools, compare sessions and use ten read-only local MCP tools. No account, Google login, bridge or cloud service is required. Findings point to evidence, not productivity scores.

Version 2.1.0. See what this release includes and the release history from 1.0.0. macOS binaries are unsigned and unnotarized; the published SHA-256 checksums verify download integrity, not publisher identity. Do not disable Gatekeeper globally.

Supported first-release platforms: macOS Intel (AMD64) and Linux AMD64 only. Apple Silicon and Linux ARM64 are deferred; no ARM download or native support is offered in this release.

1. Homebrew

Install from LanceSandino/homebrew-tap. The formula offers macOS Intel and Linux amd64; ARM builds are not offered yet.

brew tap LanceSandino/tap
brew install LanceSandino/tap/sessionsift
sessionsift doctor
brew services start sessionsift
sessionsift status
sessionsift web

The service runs the existing foreground agent as your user, not root. No sudo is needed. Alternatively run sessionsift start and leave that terminal open; do not start both on the same port. For headless use, sessionsift web --no-browser checks readiness and prints the loopback URL.

Stop and uninstall without deleting your library or source transcripts:

brew services stop sessionsift
brew uninstall sessionsift

Undo opted-in MCP configuration before uninstalling. Data remains in ~/.sessionsift; review and back up that directory before any separate deletion.

To update, stop the service and undo any configuration pinned to the old Cellar executable, then upgrade and reapply the reviewed MCP setup:

brew services stop sessionsift
brew update
brew upgrade LanceSandino/tap/sessionsift
brew services start sessionsift
sessionsift doctor

Homebrew updates the installed package; there is no sessionsift update command.

Release archive

Each release has sessionsift_<version>_darwin_amd64.tar.gz, sessionsift_<version>_linux_amd64.tar.gz, an SPDX inventory and SHA256SUMS. Verify and unpack, then run the binary:

shasum -a 256 -c SHA256SUMS --ignore-missing
tar -xzf sessionsift_2.1.0_darwin_amd64.tar.gz
./sessionsift_2.1.0_darwin_amd64/sessionsift doctor

Related MCP server: claude-replay

2. Docker

Requires a maintained Docker Engine/Desktop (Engine at least 28.5.2) and Compose v2.24.4+; see security/version requirements. From this source directory:

docker compose --env-file /dev/null -f compose.local.yaml up -d --build
docker compose --env-file /dev/null -f compose.local.yaml ps
docker compose --env-file /dev/null -f compose.local.yaml exec -T agent /app/sessionsift doctor

Open http://127.0.0.1:7777. This locally built image uses a non-root/read-only container and persistent named volume. No transcripts are mounted by default. Docker instructions cover safe optional read-only mounts, imports, Codex/Claude stdio MCP, permissions and teardown. The dedicated network is not an outbound firewall. There is no published SessionSift image to pull yet.

docker compose --env-file /dev/null -f compose.local.yaml down

This preserves data. Do not add --volumes. Do not use private/operator Compose files for consumer installation.

3. Build from Source

Use Go 1.27.2 (go.mod), Node.js 26 (tested with 26.9.0), npm and a source copy with LICENSE/NOTICE. Locked test dependencies require Node ^22.22.2 || ^24.15.0 || >=26.0.0; older generic “Node 22+” is insufficient. Dependency downloads initially require internet; app use does not. No private skills, credentials or infrastructure are prerequisites.

git clone https://github.com/LanceSandino/sessionsift.git
cd sessionsift
npm ci
npm test
npm run build
./dist/sessionsift version
./dist/sessionsift doctor
./dist/sessionsift start

Leave start running; in another terminal:

./dist/sessionsift web
./dist/sessionsift import
./dist/sessionsift sessions

Ctrl-C or ./dist/sessionsift stop stops the foreground agent. Removing the build/check-out does not remove ~/.sessionsift or transcripts. Keep required license/notices when redistributing a permitted build.

Use SessionSift

The local guide covers lifecycle, import/export files, Conversation/Timeline/Tools/Review, inspection, comparison, backups and uninstall. MCP setup includes preview, explicit safe configuration, verification and undo for local Codex and Claude Code. sessionsift setup mcp still prints configuration without modifying files. Hooks remain optional and disabled until explicitly enabled.

Default sources are Codex sessions/archives and Claude Code projects. Imports are incremental. Missing optional sources and an empty library are normal. Redaction is not a guarantee of secret-free records. AI clients you connect may send tool output to their own provider; local MCP does not make that client offline. Read the privacy model. Never expose or tunnel the Local Agent dashboard to a LAN or the internet.

Remote support, hosted MCP and encrypted cloud sync are not offered in the initial local-only release. Preserved generic bridge source uses the same binary, not another prerequisite or a promised public service. Advanced controls are hidden until explicitly requested. Basic self-host instructions apply only to deliberate use of that deferred functionality. Remote MCP is not end-to-end encrypted.

Synthetic conversation workspace

The screenshot tour contains demo data. Report vulnerabilities privately through SECURITY.md, never with real transcripts or credentials in a public issue.

License

Source-available under PolyForm Noncommercial 1.0.0 with required NOTICE attribution. Not OSI open source; commercial use is not licensed.

Related MCP Connectors

Related MCP Servers

  • A
    license
    A
    quality
    B
    maintenance
    A read-only MCP server that exposes local coding-agent session logs as three tools for introspection of recent work, debugging tool failures, and tracking token usage and estimated cost without parsing log files.
    3
    MIT
  • A
    license
    A
    quality
    A
    maintenance
    Enables search, analytics, and visualization of Claude Code sessions with MCP tools for session management, recovery, and insights.
    10
    76 PyPI
    1
    MIT
  • A
    license
    Not graded
    quality
    A
    maintenance
    Enables local-first tracking of AI token usage and costs from Claude Code and OpenCode, answering queries about tokens, models, sessions, and cost through MCP tools and a CLI, with no network calls.
    318 npm
    1
    MIT
  • A
    license
    A
    quality
    B
    maintenance
    Enables analyzing coding-agent session transcripts to measure tool efficiency, detect loops, track test outcomes, and report costs, with MCP tools for listing sessions, analyzing sessions, finding loops, and generating cost reports.
    4
    MIT