What's new in this release
This is the first release of MiroFish adapted for internal use at Slater Consulting.
Changes from upstream (666ghj/MiroFish)
English localization
- Full UI, prompts, logs, and source code translated from Chinese (60+ files)
Slater Consulting brand UI
- Dark navy theme matching Slater Consulting visual identity
- Geist Sans typography, custom color tokens throughout the Vue frontend
Multi-provider LLM support
- OpenAI, Anthropic, Claude CLI, and Codex CLI — configurable via
.env - No Alibaba/Qwen dependency
Local graph storage
- Embedded KuzuDB replaces hosted graph service — runs fully offline, no external accounts needed
Configurable rate limiting (new in v1.0.0)
- Automatic 429 retry with exponential backoff (base 30s, cap 300s, configurable retries)
- Token bucket RPM/TPM throttling — proactive enforcement before each LLM call
- Inter-turn delay injection in all simulation scripts
- Rate limit settings panel in the UI — adjust before each simulation run, persisted via localStorage
POST /api/simulation/{id}/configendpoint to write settings tosimulation_config.json