Roadmap

Aludel's roadmap follows the major product themes visible in the repository history and changelog. This page preserves the original milestone intent while distinguishing shipped work from future direction. Current issue and milestone status on GitHub is authoritative.

Recently Shipped

The current main branch has already landed several capabilities that were previously future-facing in older roadmap notes:

• App callback execution for routing runs and suites through a host application's real workflow
• json_deep_compare assertions with partial-match scoring and average suite-run scores
• Raw JSON exports for run and suite results
• Result copy actions plus support for nullable callback metrics and metadata
• Configurable document storage backends for local, AWS S3, and Google Cloud Storage
• Reusable single-turn and multi-turn datasets with provenance and idempotent suite population
• Rolling dashboard analytics, evolution deltas, regression/stability signals, and Pareto analysis
• Failure-grounded prompt suggestions with explicit acceptance and dismissal
• xAI, Groq, and OpenRouter provider support alongside OpenAI, Anthropic, Google Gemini, and Ollama
• Deterministic demo coverage across all major workflows

For the current product surface, see Home. For active work, use GitHub milestones and open issues.

Milestone Overview

The sections below are the historical plans attached to named milestones. Many of their tracked issues have since shipped; the links retain their status and discussion.

v0.1 Apprentice

The Cauldron Awakens
Due: April 13, 2026

Focus: immediate workflow cleanup and provider UX polish.

Tracked issues:

• #59 extract suite editor workflow out of SuiteLive.Show
• #58 apply prompt search and tag filters before pagination
• #30 implement custom select dropdown component
• #11 auto-populate model dropdown based on selected provider

Expected outcome:

• cleaner suite and prompt management flows
• less LiveView complexity in the editor surface
• lower-friction provider configuration
• a more polished first-use experience

v0.2 Herbalist

Roots of the Ancient Grove

Focus: architecture, supervision, run execution, and platform reliability.

Tracked issues:

• #61 architecture: introduce application-service boundaries and thin LiveViews
• #60 move run execution orchestration into a supervised application service
• #62 model explicit run execution states and transitions
• #89 recover SuiteLive when background suite execution task crashes
• #51 optimize LiveView run updates with selective fetching
• #50 make LLM request concurrency configurable
• #44 retry individual test cases
• #31 default pricing with override support
• #28 custom storage backends for test case documents
• #88 enforce assertion validation consistently across visual and JSON editors
• #87 make json_field assertions type-aware for scalar comparisons

Expected outcome:

• clearer application boundaries
• supervised and explicit execution ownership
• better UI feedback for long-running work
• stricter and more trustworthy assertion semantics
• expanded core provider coverage
• stronger production-readiness for pricing and document storage

v0.3 Alchemist

The Transmutation Begins

Focus: richer evaluation fidelity and better result inspection, building on the callback execution and structured-output work already shipped on main.

Tracked issues:

• #69 embedded app callback execution for high-fidelity evals
• #43 deep comparison scoring for structured output assertions
• #73 add copy actions for run and suite results
• #77 export raw run and suite results as JSON
• #75 support nullable metrics and callback metadata in result views

Expected outcome:

• tighten and extend the recently shipped callback-mode and structured-output workflows
• improve debugging and result reuse around exported artifacts and copied outputs
• keep result views trustworthy when execution paths return uneven telemetry

v1.0 Archdruid

The Great Work is Done

Focus: first-class workflow evaluation.

Tracked issues:

• #42 workflow evaluation support

Expected outcome:

• multi-step workflow test cases
• workflow-level cost and latency tracking
• richer comparison of real-world LLM application behavior

v1.1 Sage

The Knowing Deepens

Focus: decision-oriented analytics and regression visibility.

Tracked issues:

• #78 rework dashboard stats around rolling windows and period deltas
• #81 add outcome-weighted efficiency metrics
• #80 add version-over-version deltas to prompt evolution
• #79 add regression and stability signals to analytics

Expected outcome:

• analytics that stay useful after long-term use
• clearer tradeoff metrics tying quality to cost and latency
• better visibility into regressions and instability
• stronger prompt and provider decision support

v1.2 Oracle

What Was Hidden, Speaks

Focus: richer prompt analysis surfaces after analytics and workflow primitives are in place.

Tracked issues:

• #76 export prompt evolution metrics
• #74 add prompt version diff view

Expected outcome:

• clearer prompt history analysis and comparison
• exportable metrics for external analysis
• better visibility into how prompt iterations change outcomes

v2.0 Hierophant

The Sacred Patterns Revealed

Focus: optimization and recommendation systems built on top of mature evaluation and analytics.

Tracked issues:

• #12 integrate GEPA for prompt evolution and optimization

Expected outcome:

• prompt evolution intelligence
• optimization-oriented workflows on top of trusted evaluation history
• a clearer path from measurement to automated improvement

Unassigned Future Milestones

The following milestones exist but currently have no open issues assigned:

• v0.4 Druid
• v2.1 Magus
• v3.0 Aetherius

These are good candidates for:

• polish and consolidation
• platform scale features
• advanced workflow orchestration
• long-horizon optimization and recommendation systems

Strategic Summary

1. Refine the immediate UX and trim workflow-heavy LiveViews.
2. Strengthen execution ownership and platform reliability.
3. Improve evaluation fidelity, structured scoring, and result inspection.
4. Expand into workflow evaluation.
5. Make analytics decision-oriented rather than cumulative.
6. Improve prompt analysis and comparison surfaces.
7. Build optimization intelligence on top of mature evaluation primitives.