Coding Agent Session Search (cass) — Indexes and searches your local AI coding agent session history across 30+ providers.
Analyzed by Sai Pavan Gopularam · AI · Developer Tools · View on GitHub
- Stars: 1200
- Forks: 146
- Commits last 30 days: 100
- Health: Active (100 commits this month)
- Language: Rust
- License: Other
What It Is
Imagine a universal library for all your AI coding assistant chats. `cass` (coding-agent-search) is a command-line tool and TUI that acts like a personal librarian, collecting every conversation you've had with tools like Copilot, Claude, Gemini, and 30+ others. It normalizes their disparate formats into a single, searchable timeline.
This kills the problem of fragmented knowledge where valuable solutions and debugging steps are lost across different agent interfaces. Instead of scrolling endlessly or forgetting past insights, `cass` makes your entire coding history instantly retrievable, even understanding natural language queries.
License Verdict
Non-Standard License — Proceed with Caution — Commercial Use Uncertain • Review 'MIT+OpenAI/Anthropic Rider'
The project uses a non-standard 'MIT+OpenAI/Anthropic Rider' license. While MIT is permissive, the 'Rider' implies additional, unspecified terms from OpenAI and Anthropic. Founders must carefully review these riders to understand any potential restrictions on commercial use, redistribution, or modification before building a business on this codebase.
How to Use It
Install `cass` via a shell script or package manager. Once installed, it automatically indexes your local agent session files. You can then use its TUI or CLI to search and retrieve past conversations.
Prerequisites:
- Rust (for source install)
- Linux | macOS | Windows
Estimated setup time: 5 minutes.
curl -fsSL "https://raw.githubusercontent.com/Dicklesworthstone/coding_agent_session_search/main/install.sh" | bash -s -- --easy-mode
cass search "how to fix memory leak" --limit 3
cass --version
What I'd Build With This
Unified Agent Chat Archive & Export (micro-saas)
Offer a simple web service where developers can upload their raw agent session files (JSONL, SQLite, etc.) from various tools. The service processes and archives them, providing a unified, searchable view and allowing export to common formats (Markdown, PDF, JSON). It solves the pain of manual archiving and cross-agent search for individual developers.
Effort: 1 Week Build Time · Target: Individual Developers, Freelancers · Pricing: $9/month per user
Team AI Agent Knowledge Base & Collaboration (saas)
Build a collaborative SaaS platform for development teams. Team members can securely connect their local `cass` instances or upload their agent histories to a central, private knowledge base. This allows for shared search, discovery of team-specific solutions, and reduces redundant problem-solving. Features could include project-based organization, user roles, and semantic search capabilities.
Effort: 3 Months Build Time · Target: Small to Medium Dev Teams · Pricing: $49/month per team of 5
AI Agent Interaction Audit & Governance Platform (enterprise)
Develop an on-premise or private cloud solution for large enterprises to centralize and audit all AI agent interactions across their developer workforce. This platform would ensure compliance with internal policies, identify potential data leakage, and provide deep analytics on agent usage and effectiveness. Custom connectors for enterprise-specific data sources and reporting dashboards would be key.
Effort: 6 Months Build Time · Target: Large Enterprises, Regulated Industries · Pricing: $5,000+/month custom contract
Sai Pavan Gopularam's Take
This tool is a goldmine for developers drowning in agent chats. The core value is unifying fragmented knowledge, which is a common problem. I'd lean into offering a hosted version for individual developers, charging around $15/month for unlimited storage and advanced search, making it a no-brainer for pros.
Watch Out For
- Non-Standard License: The 'MIT+OpenAI/Anthropic Rider' license is not standard. Founders must investigate the 'Rider' terms carefully as they could impose commercial restrictions or obligations beyond typical MIT terms.
- Agent Mode Caveats: When using `cass` in an automated agent context, always use `--robot` or `--json` flags. Running `cass` bare will launch the interactive TUI, which is not suitable for programmatic use and can cause unexpected behavior.
- Semantic Search Opt-In: Semantic search capabilities require an explicit `cass models install` command to download local models. Until models are installed, `cass` will fall back to lexical-only search, which may not provide the same depth of natural language understanding.
- External Interruptions: If an external process interrupts `cass` (e.g., using GNU `timeout`), it may leave incomplete JSON output. Always inspect the `budget.timed_out` field even on exit 0 to ensure results are complete.