📝 Changelog

[0.2.0] - 2025-07-30

✨ Added

• 🧠 Multi-backend conversation support Added support for persistent memory across multiple LLM backends using a new architecture. Enables PostgreSQL-based WAL persistence.

• 🔮 Google Gemini support Introduced support for Gemini LLMs and embedding models.

• ☁️ Alibaba Cloud Qwen support Added compatibility for Qwen models.

• 🧬 Gemini & Ollama embedding providers Cipher now supports embedding generation via Google Gemini and Ollama APIs.

• 🛡️ Embedding fallback mechanism Implements automatic fallback logic for embeddings. If the primary provider is unavailable, Cipher selects the next available provider based on environment variables.

• 📦 Prompt Provider Support Added extensible system prompt architecture with dynamic, static, and file-based providers. Enables customizable prompt injection through CLI commands (/prompt-providers) with support for conversation summaries, project guidelines, and real-time prompt management.

• 📊 Token Management Implemented intelligent token counting and context compression with provider-specific tokenization. Features automatic compression when approaching context limits, token usage statistics, and configurable compression strategies for optimal memory management across different LLM providers.

• 🧩 Aggregator Mode support Implements Aggregator mode for Cipher's MCP server, which exposes all Cipher's tools to agents/clients.

📚 Documentation

• 🚀 New example: MCP aggregator hub (Use Case 4) Introduced a new real-world example in examples/usecase-4, showing how Cipher can aggregate multiple MCP streams.

• 📝 Improved README and prompt documentation Updated descriptions, prompt templates, and environment variable explanations to reflect the latest architecture and provider support.