Prism

An explainable decision engine for model routing.

Prism separates routing from execution, allowing applications to choose the right model through explicit capabilities, policies, and transparent decisions.

---

Why Prism?

Modern AI applications rarely rely on a single model.

As providers and models grow, routing logic gradually spreads throughout the application:

• provider-specific conditionals
• capability checks
• latency heuristics
• cost thresholds
• fallback chains

What begins as a few if statements eventually becomes infrastructure that is difficult to reason about, extend, and test.

Routing is a system of its own. Prism treats it that way.

---

How It Works

Instead of asking which provider should I call?, applications describe what they need.

Prism then:

1. analyzes the request
2. identifies the required capabilities
3. evaluates routing policies
4. selects the best execution target
5. explains the decision

Execution is optional. Prism can recommend a model or execute the request through provider adapters.

---

Crates

Crate	Description
prism-core	Domain types, traits, and contracts shared across all crates
prism-engine	13-stage deterministic routing pipeline
prism-registry	Model profile storage and retrieval
prism-tui	Terminal interface for interactive decision exploration

prism-tui

Launch with cargo run to enter an interactive prompt evaluator.

Type or paste a request, press Enter, and explore the full routing pipeline:

• pipeline sidebar with stage-by-stage navigation
• structured workspace views per stage
• contextual details panel
• search, replay, and JSON export

---

Core Ideas

Routing is independent

Model selection should not be coupled to application logic.

Decisions should be explainable

Every routing decision should be inspectable and reproducible.

Policies drive routing

Cost, latency, privacy, governance, and organization-specific rules determine how routing decisions are made.

Providers are interchangeable

Applications depend on capabilities rather than vendor-specific APIs.

---

Pipeline

Prompt
  ↓
Normalization
  ↓
Intrinsic Extraction  →  Derived Analysis
  ↓
Requirement Inference  →  Capability Mapping  →  Prioritization
  ↓
Candidate Filtering
  ↓
Policy Evaluation
  ↓
Candidate Scoring
  ↓
Decision Selection
  ↓
Explanation Generation
  ↓
Decision Report

---

Roadmap

• Core routing engine
• Capability taxonomy and extraction
• Policy framework with typed rules
• Candidate scoring with evidence
• Interactive terminal UI
• Provider adapters
• HTTP API
• Configuration DSL
• Plugin system

---

Getting Started

cargo build
cargo test
cargo run -p prism-tui

---

Contributing

We welcome contributions from the community.

See CONTRIBUTING.md for setup instructions and contribution guidelines.

---