agentlore

Agentlore

Open a PR, see the AI conversation that produced it. Agentlore turns your team's agent conversations into a shared, searchable resource linked to your code.

Why

AI coding agents are part of everyday development, but conversations stay trapped on individual machines. Agentlore makes them a shared, searchable team resource — linked to your code.

How It Works

1. Developers use their agents as usual — Claude Code, Cursor, Copilot, Gemini CLI, Codex, Amp, OpenCode. Nothing changes in the workflow.
2. agentsview collects local sessions into a SQLite database on each developer's machine.
3. The agentlore daemon watches for new sessions, masks secrets (API keys, credentials), and syncs conversations to shared ClickHouse storage.
4. The web UI lets the team browse conversations, full-text search across all sessions, and navigate from git commits/PRs back to the conversations that produced them.

Git linking works by parsing tool calls in conversations for git commit, git push, and gh pr create invocations, extracting commit SHAs and PR references.

Prerequisites

Install agentsview to collect local agent sessions:

curl -fsSL https://raw.githubusercontent.com/wesm/agentsview/main/install.sh | bash

See the agentsview README for other install methods.

Install

git clone https://github.com/clkao/agentlore.git
cd agentlore
make build install    # builds frontend + Go binary, installs to ~/.local/bin

Requires Go 1.25+, Node 22+, and Docker (for ClickHouse).

Start ClickHouse:

docker compose up -d

Quick Start

# Start agentsview to collect local sessions (runs on port 8080)
agentsview &

# Sync conversations to ClickHouse and start the web UI
agentlore sync
agentlore serve
# Open http://localhost:9090

Configuration

Agentlore reads ~/.config/agentlore/config.json. All fields are optional — defaults work for local use with docker compose.

{
  "clickhouse_addr": "localhost:9000",
  "clickhouse_database": "agentlore",
  "clickhouse_user": "",
  "clickhouse_password": "",
  "agentsview_db_path": "~/.agentsview/sessions.db",
  "server_port": 9090
}

For remote or managed ClickHouse, set clickhouse_addr, clickhouse_secure, and credentials. The agentsview_db_path defaults to ~/.agentsview/sessions.db or ~/.claude/agentsview/sessions.db, whichever exists.

Current Limitations

• ClickHouse full-text search requires v26+ — the FTS features agentlore uses are not yet available in ClickHouse Cloud.
• Some tool results are not stored — agentsview filters out verbose tool outputs (Read, Glob, etc.) by default, so those won't appear in synced conversations.
• Git commit linking is fragile — SHA extraction from tool calls works for common patterns but can miss or misattribute commits in edge cases.
• Syncs everything agentsview sees — there's no per-project filter in agentlore itself. To limit scope, create a custom projects directory with symlinks to the sessions you want indexed, then point agentsview at it:

  mkdir -p ~/agentlore-projects
  ln -s ~/.claude/projects/-Users-you-work-myproject ~/agentlore-projects/
  CLAUDE_PROJECTS_DIR=~/agentlore-projects agentsview -no-browser
  

  Then set agentsview_db_path in agentlore's config to use that database.

Roadmap

• Improve git commit and PR linking reliability
• Project identity reconciliation (map paths to canonical project IDs after the fact)
• Guides for setting up auth (Cloudflare Access, etc.)
• agentsview insights sharing

Development

# Prerequisites: Go 1.25+, Node 22+, Docker

# Backend
go test ./internal/...

# Frontend
cd frontend && npm install && npm run dev

# Full build (frontend + Go binary)
make build

Acknowledgements

• agentsview by Wes McKinney — the local agent session collector that agentlore reads from.
• ClickHouse — the storage engine powering conversation search and analytics.

License

MIT