CoreMind Public Roadmap

CoreMind helps newcomers and application engineers build reliable business agents through configuration, one standard runtime, a bounded harness and loop, quality gates, and task-oriented learning materials.

This public roadmap describes product boundaries and intended directions. It deliberately excludes internal schedules, acceptance records, and maintainer working notes. Priorities may change as real users and community contributors provide feedback.

Current stable release: 0.3.0 (published)

0.2.0-rc.1 remains the immutable reference baseline. GitHub Releases, npm, and PyPI determine the latest installable version in each channel. The current 0.3.0 stable release retains these rc.2 capabilities:

• Three use paths: CLI/TUI, TypeScript and Python SDKs, and full source.
• Single-agent and multi-agent execution, workflows, and budget-bounded loops.
• Config v2, 40 configurable model providers, and custom compatible endpoints.
• ask, assisted, and full permission modes with path policy, approvals, audit, checkpoints, diffs, and recovery.
• Explicit outcomes, budgets, traces, sessions, context protection, tests, evaluations, and release gates.
• Eight project templates, five offline golden examples, two real-defect coding-agent repositories, and 21 capability modules.
• Tests, SOPs, a Skill, bilingual guides, and examples for every capability module.
• Acceptance workflows that combine Windows/Linux automation with real pseudoterminal checks on both platforms, plus synchronized GitHub, npm, PyPI, and bilingual documentation releases.

0.3.0 is publicly released. Controlled dogfooding of the rc.2 CLI, TypeScript SDK, Python SDK, independent source package, and Windows TUI found no stable-release blocker. The seven-check live-provider revalidation, both-platform automated gates, exact-main identity, final maintainer acceptance, and Release Readiness are complete.

It includes the public loop configuration, verify-repair states, stable snapshots, pause-resume, effect receipts, bounded retries, a fifth verified-repair golden example, plus the controlled process, read-only Git, bounded unified diff, seven grader types, and TypeScript/Python real-defect evaluations required by coding agents. Every candidate must complete the automated quality gates, both-platform P01-P20, a current live-provider recheck, and the final documentation audit on the same commit before synchronized publication.

0.3.x: current stabilization

The current stabilization line focuses on the reliability and usability of the capabilities delivered in 0.3.0-rc.2 in the 0.3.0 stable release and later 0.3.x updates:

• Continue improving the Windows and Linux TUI experience and terminal compatibility.
• Fix installation, configuration, diagnostics, and interaction issues found during initial community use.
• Expand provider certification backed by real invocation evidence.
• Continue cross-language checks for CLI, TypeScript SDK, and Python SDK outcomes and events.
• Continue including bounded-Loop, harness, and coding-agent evidence in cross-platform release-candidate acceptance.
• Strengthen security, recovery, evaluation, documentation, and public-package release regression coverage.

A provider that has not passed live verification remains configurable but is not marked as officially certified.

0.3.0: phase-two kernel and engineering loop

Phase two advances through alpha → beta → rc and strengthens the framework kernel without changing CoreMind's three use paths:

• Align critical dependency versions and isolate low-level implementations behind private adapters.
• Establish durable harness operations that can abort, pause, resume, and recover without duplicating committed effects.
• Improve context selection, compaction evidence, long-output artifacts, and observable metrics.
• Build a first-party Coding/Engineering Kernel for repository understanding, minimal edits, diffs, verification, repair, and regression.
• Keep outcomes, events, approvals, and recovery consistent across TUI, headless CLI, TypeScript SDK, and Python SDK.
• Ship tests, SOP, Skill, bilingual guides, examples, migration, and rollback guidance with every batch.

0.3.0-rc.2 completed Batches 0 through 6: critical dependencies use one exact family behind private adapters; runs have a durable operation envelope, atomic RunState, Session backend conformance, backed-up migration, and non-replay boundaries for uncertain effects; long tasks add stable context prefixes, deterministic compaction evidence, truthful cache metrics, and controlled workspace artifacts; coding is a first-party Engineering Kernel inside the Runtime; and the extension surface is limited to four controlled lifecycle events with traceable experiments, a seven-check provider contract, TUI evidence views, and one shared RunSnapshot across all entry points. The current 0.3.0 candidate adds no product behavior and only synchronizes stable versions and release materials. The current evidence ledger, Releases, and registries remain authoritative for candidate provider status, public availability, and final assets.

Phase two starts from an immutable 0.2.0-rc.1 reference baseline that freezes public types, Config/Protocol schemas, the critical dependency combination, P01-P20, both-platform behavior, same-task coding-evaluation conditions, and quality floors. Coverage may improve but may not regress. Every intentional contract change must explain migration and rollback; gates are never lowered merely to pass.

Phase three: Web development environment

Phase three is planned to add a complete browser-based development experience:

• Visual configuration of agents, tools, and workflows.
• Online code editing and project file management.
• Trace debugging, testing, and evaluation panels.
• Permission approval, run-state, and result inspection.
• Release and deployment guidance.

The Web environment will reuse CoreMind Protocol and the existing runtime instead of introducing another execution engine.

Later platform and ecosystem work

• Formal macOS support.
• More community templates, Skills, business modules, and golden examples.
• Live certification for additional model providers.
• Continued improvements to contributor development, testing, review, and release workflows.

Every new capability must continue to ship with implementation, tests, SOP, Skill, bilingual guidance, and examples.

Explicit boundaries

CoreMind does not decide business goals, data fields, approval ownership, or agent architecture for users, and it does not promise to implement every business application automatically. There is currently no plan for an official hosted API, multi-tenant SaaS, official Docker image, or separate pure-Python runtime.

Users own business direction and final acceptance. CoreMind owns engineering mechanisms, safety boundaries, quality evidence, and development guidance.

Participate in the roadmap

Use GitHub Issues for adoption feedback, defects, and capability proposals. Before contributing code, read the contribution guide and security policy.