M.I.M.I.R - Multi-agent Intelligent Memory & Insight Repository

# Changelog ## [0.1.0] - 2025-11-19 ### Added - **Model Selection Support**: Accept `model` parameter in chat requests (OpenAI-compatible) - Configure via `mimir.model` setting (default: `gpt-4.1`) - Supports any model available in Mimir: GPT-4, Claude, Llama, etc. - Model is preserved across conversation (unless explicitly changed) ### Fixed - **Chat History Handling**: Proper context preservation across conversation - System message (preamble) is only added once per conversation - Follow-up messages reuse the original preamble automatically - Slash commands can override preamble mid-conversation - No more duplicate system messages - **Tool Parameters**: All tool configurations preserved across messages - Vector search depth/limit/similarity persist throughout conversation - Tool enable/disable settings maintained per chat thread ### Improved - **Chat Continuity**: Better conversation flow - Starting with `/research` keeps research mode for all follow-ups - Model selection persists (from settings) - All configurations apply to entire conversation unless changed ## Usage Examples ### Basic Chat ``` @mimir how does graph-rag work? ``` Uses: default preamble (`mimir-v2`) + default model (`gpt-4.1`) ### With Slash Command ``` @mimir /research what are Neo4j best practices? ``` Uses: `research` preamble + default model + higher search depth ### Follow-up Messages ``` @mimir /debug why is my API failing? @mimir can you check the logs? ``` Both use: `debug` preamble (preserves from first message) + same model ### Change Model Mid-Conversation Update `mimir.model` in settings to switch models (takes effect next message) ## Configuration Reference | Setting | Default | Description | |---------|---------|-------------| | `mimir.model` | `gpt-4.1` | LLM model to use | | `mimir.defaultPreamble` | `mimir-v2` | Default chatmode | | `mimir.vectorSearch.depth` | `1` | Graph traversal depth (1-3) | | `mimir.enableTools` | `true` | Enable MCP tools | | `mimir.maxToolCalls` | `3` | Max tool calls per response |