v0.1.0-beta.1
Releasing soon — First public beta This is the first public beta release. The full feature set is complete and tested. Feedback from real-world usage will drive patches (0.1.0-beta.2, etc.) before the 1.0.0 stable GA release.
Report issues at github.com/satyendra2013/mira-app/issues
v1.0.0 (upcoming stable release)
Core application
- Electron 28 desktop app for macOS, Windows, and Linux
- Bundled Python 3.11 runtime — no system Python required
- SQLite-backed local storage — all data stays on your machine
- OS keychain integration for secure API key storage
- Dark, light, and system theme support
Reasoning engines
- Native Agent Engine (NAE) — multi-layer context management, token ledger, parallel sub-agent spawning (up to 4 sub-agents, parallelism 2)
- RLM Engine — iterative Python code-writing and execution loop, configurable max iterations (default: 30), early-stop on solution detection
- Supported providers: OpenAI, Anthropic, AWS Bedrock, Ollama
Features
- Skills — 5 built-in domain personas; custom skill creation with model overrides; session and global scope; JSON import/export
- Workflows — 3 built-in pipelines; custom builder with prompt, transform, and human_review step types; template variables; routing rules; JSON import/export
- Document context — PDF, DOCX, TXT, MD, CSV, JSON, and code files; fixed-size chunking; up to 50 MB per file
- MCP Tools — stdio and SSE transport; session and global scope; per-server tool allowlisting
- Eval Framework — exact match, semantic similarity, LLM judge, and human review eval types; run history; A/B comparison; CSV and JSON export
- REPL Console — real-time visibility into every reasoning step
- Activity Center — full session history and audit trail