Empathy Framework is an AI-driven developer workflow framework whose core goal is to transform AI from a passive responder into a collaborative partner with predictive capabilities. It combines a five-level AI framework with git-native memory (MemDocs) to overcome session amnesia, and claims it can anticipate problems weeks in advance.
Purpose and Positioning The framework is aimed at developers and provides code review, debugging, testing, and release workflows run from the terminal or Claude Code. Its intelligent tiered routing can save 34-86% of LLM costs. The project is transitioning to a Claude-native architecture to unlock capabilities that multi-provider abstraction cannot achieve.
Main Capabilities
- Workflows: code review, debugging, test generation, release preparation, security audit, performance audit, documentation generation, defect prediction, etc.
- Cost optimization: intelligent tiered routing, response caching (up to 57% cache hit rate for similar prompts), batch API (50% cost savings), precise token counting (>98% accuracy), cache performance monitoring.
- Multi-agent orchestration: supports custom agents and Anthropic LLM agents, provides an agent coordination dashboard, including 6 coordination modes such as heartbeats, coordination signals, event streams, approval gates, and quality feedback.
- Authentication policy system: intelligently routes between Claude subscriptions and the Anthropic API, using subscriptions (free) for small/medium modules (<2000 LOC) and the API for large modules.
- MCP server integration: exposes workflows as native Claude Code tools, including 10 tools in total: security_audit, bug_predict, code_review, test_generation, performance_audit, release_prep, auth_status, auth_recommend, telemetry_stats, and dashboard_status.
- Natural language workflows: supports describing requirements in natural language and automatically matching workflows.
- Socratic workflows: interactive discovery through guiding questions, dynamically creating agents.
- Pattern learning: workflows learn from results and improve over time.
- Security: path traversal protection, JWT authentication and rate limiting, PII scrubbing in telemetry, HIPAA/GDPR compliance options, automated security scanning (82% accuracy).
Usage
- Installation: installation options and environment setup instructions are provided (specific commands are not given in the materials).
- VSCode extension: provides a dashboard, one-click workflows, a memory panel, and cost tracking.
- In Claude Code: through MCP server integration, tools are automatically discovered without manual configuration (via .claude/mcp.json).
- Dashboard: accessible via python examples/dashboard_demo.py http://localhost:8000（需要 Redis).
- Workflow organization: divided into command centers such as /workflows (automated analysis) and /plan (planning/review).
Applicable Scenarios
- Developers' daily code review, debugging, testing, and release processes.
- Security scanning and cost optimization in CI/CD automation.
- Complex workflows requiring multi-agent collaboration and human approval steps.
- Teams using Claude subscriptions or the Anthropic API that want to reduce LLM costs.
Limitations and Notes
- Since v5.0.0 (January 26, 2026), non-Anthropic providers (OpenAI/Google/Ollama) have been removed, which is a breaking change.
- The dashboard requires Redis support.
- Some features (such as MCP integration and authentication policy) have version requirements (v5.1.1+, etc.).
- Details such as specific installation commands, pricing, and platform support are not yet provided in the materials.
- The project uses the Apache License 2.0 and can be used, modified, and used to build commercial products for free.
