Summary

seCall addresses the challenge of managing and recalling information scattered across AI agent conversations. It provides a local-first solution for collecting, organizing, and searching these interactions, transforming them into a searchable local wiki. This benefits engineers and researchers by creating a centralized, easily retrievable knowledge base of AI-driven development and decision-making.

How It Works

seCall ingests session logs from multiple AI agents (Claude Code, Codex CLI, Gemini CLI, claude.ai, ChatGPT) into a unified format, storing transcripts and AI-generated wiki content in an Obsidian-compatible Markdown vault. It features a hybrid search combining BM25 (SQLite FTS5 with Korean NLP) and vector semantic search (Ollama BGE-M3 embeddings, usearch HNSW), fused via RRF. Configurable LLM backends generate wiki content and expand queries. A knowledge graph built from session relationships supports the MCP protocol. Access is via Web UI, REST API, and Obsidian plugin, all sharing a common API.

Quick Start & Requirements

Installation via pre-built binaries from GitHub Releases is recommended. Source builds require Rust 1.75+ and optionally Node.js 22/pnpm 9 for the web UI. Prerequisites include an AI agent (Claude Code, Codex CLI, Gemini CLI) and optionally Ollama for vector search/LLM backends. Windows users need the MSVC toolchain.

Highlighted Details

• Multi-Agent Ingestion: Parses sessions from popular AI coding assistants.
• Hybrid Search: Combines BM25 (SQLite FTS5) and vector semantic search (Ollama BGE-M3, HNSW) with RRF and LLM query expansion.
• Obsidian Vault: Organizes data into a two-tier Markdown vault (raw sessions, AI wiki) with full Obsidian compatibility.
• Knowledge Graph: Builds a graph of session relationships, supporting MCP protocol.
• Multi-Device Git Sync: Synchronizes the knowledge vault across machines.
• Unified Interfaces: Offers Web UI, REST API, and Obsidian plugin sharing a common API.

Maintenance & Community

Actively maintained with frequent updates detailed in its changelog. Direct contact is via email (d9ng@outlook.com). No explicit community channels (Discord, Slack) are listed.

Licensing & Compatibility

Licensed under AGPL-3.0. This strong copyleft license requires distributed modifications to be open-sourced under the same terms, potentially restricting integration with closed-source commercial products.

Limitations & Caveats

On Windows, HNSW ANN indexing and Kiwi-rs morphological analysis are unavailable, falling back to alternatives. The AGPL-3.0 license necessitates careful consideration for commercial closed-source applications.