Product Roadmap

Priorities are driven by user feedback and the actual needs of prompt engineering.
This plan is a living document and will evolve as we build.

Available Now

v1.0 Public Release

Our official stable launch. A clean, debt-free architecture ready for production workloads.

Multi-Model Shotgun: Route 1 prompt to N providers simultaneously.
Visual Comparison: Side-by-side result grid for quick iteration.
Prompt Variables: Support for templating (e.g., {{name}}) for dynamic testing.
Model Evaluations: Success criteria and manual output grading to score model results.
Parameter Control: Per-model adjustment of Temperature, Top_P, and Stop Sequences.
History Explorer: Browse, search, and revive past sessions.
Encrypted Storage: API Keys stored in your local OS keyring.

Just Shipped

Major upgrades to evaluation analytics, artifact exports, and strict schema enforcement.

Structured Outputs: Enforce strict JSON Schema generation natively across OpenAI, Anthropic, Mistral, and Gemini.
Visual Dashboard: Comprehensive charts for Latency vs Pass Rate, Accuracy, and Token usage.
True Throughput Tracking: Isolate engine speed by calculating exact Output Tokens per Second.
Data Exports: Download presentation-ready PNG charts and comprehensive CSV leaderboards.
Split-Context Editor: A new modular block-based editor for writing prompts alongside JSON schemas.

Up Next: v1.2

Expanding the Providers Vault to support self-hosted and local models.

Local Server Support: Add custom OpenAI-compatible endpoints to connect directly to local instances of llama.cpp, vLLM, Ollama, and LM Studio.

Planned: v1.3

Adding static pricing data to provide cost context during testing.

Evaluation Cost Metrics: Attach pricing information to the LLM, to automatically estimate the API cost of bulk evaluation runs based on token usage.

AptSelect is driven by the actual needs of prompt engineering. What should we prioritize next?