KMSES
In developmentAn autonomous multi-agent engineering system. CrewAI + Celery + Redis + fine-tuned LLMs that build, test, and ship code across the KAKEL portfolio.
Inquire about licensingAbout
KMSES (Kakel Multi-agent Software Engineering System), internally called ai-company, is the engineering autopilot KAKEL uses to keep dev velocity high across a multi-vertical portfolio. CrewAI agents coordinate over Celery + Redis; each agent has a fine-tuned LLM profile (OpenAI fine-tunes for domain-specific code generation) and a tool kit. Architectural pieces include a planner, a coder, a reviewer, a test-author, and a release-engineer. Persistent memory in ChromaDB. LiteLLM provider abstraction so any agent can swap to a different model. Currently internal-facing only. External licensing being scoped.
Features
- — CrewAI multi-agent orchestration with role specialization
- — Celery + Redis task graph for parallel agent execution
- — Per-agent fine-tuned model override (OpenAI fine-tunes via LiteLLM)
- — ChromaDB persistent memory across runs
- — Ollama support for on-prem inference
- — Engineering output telemetry: PRs opened, tests passed, regressions caught