[ ai dev environment ]
Zenith
An AI-native developer environment.
Most AI coding tools revolve around a conversation between one developer and one model. Zenith explores reasoning models coordinating specialized engineering agents.
- Developer
- Reasoning model
- Plan
- Code agent
- Audit agent
- Research agent
- Results
- Reviewer
- Developer
A powerful reasoning model holds the objective while smaller, specialized models execute bounded tasks — so expensive reasoning is spent only where it actually adds value.
Codebase-level context. Agents explore entire repositories rather than isolated files: architecture discovery, dead code detection, performance audits, modularity and dependency analysis, bug investigation, refactoring planning, test coverage. The point is giving agents enough context to reason about systems, not snippets.
Observability-aware engineering. The area I'm most interested in: connecting coding agents to production telemetry instead of source code alone.
- P99 latency increased
- Agent inspects traces
- Finds slow database path
- Maps trace to source code
- Proposes optimization
- Validates against architecture
Local and open-weight models. Zenith doubles as an environment for experimenting with local inference, model routing, Mixture-of-Experts architectures, quantization, speculative decoding and heterogeneous GPU/CPU inference. The goal is not depending on a single model or provider.
Stack Rust · GPUI · TypeScript · LLM APIs · open-weight models · MCP · local inference · agent orchestration