Skip to content

NextClaw Roadmap

NextClaw aims to become a proactive, continuously improving, long-term AI partner, with one fundamental mission: to be a good partner to its user. See the Vision.

The goal for 1.0: deliver a long-term AI partner that understands you, takes useful initiative, and continuously improves itself.

The four stages below describe planned work toward 1.0. Full acceptance testing is not complete, and no release date is set. Development will build on existing memory, scheduling, conversations, skills, and result delivery; these are neither all-new systems nor capabilities that are all available today. See Release Notes for shipped features.

Stage 1: Understand You and Your Goals

Help your partner keep track of your background, goals, and current circumstances, so a new conversation can still use the latest relevant information.

Planned featureWhat it will let you do
Partner identity and ongoing participation settingsDecide what it may proactively participate in, which material it may use, and which actions it may take; turn initiative off at any time
Automatic organization of user informationBuild an understanding of your background, preferences, and circumstances through normal conversation, distinguishing confirmed facts from interpretations that need confirmation
Review, correction, and expiry of remembered informationInspect its understanding, correct mistakes or stop using a fact, and reconfirm temporary circumstances when they expire
Personal goals and plans across time horizonsManage long-term goals, intermediate plans, near-term actions, priorities, and progress, including goals that do not belong to a project
Relevant context across conversations and background workUse current, relevant, authorized background in new conversations and background analysis, with an explanation of what informed a suggestion

Stage 2: Proactive Analysis, Planning, and Conversation

Without a new request, your partner should be able to revisit existing information at an appropriate time, share a relevant change, or reach out with something worth discussing.

Planned featureWhat it will let you do
Automatic identification and management of ongoing interestsKeep track of goals, circumstances, and unresolved questions; see why something matters, recent results, and follow-up arrangements, with pause and stop controls
Background analysis and planningNotice conflicts between goals and plans, compare options, and offer suggestions; gain useful new insights even when nothing external has changed
Timing and frequency controlsSet quiet hours, frequency, and budget; stay quiet when there is no new value, avoid repeated reminders, and pause initiative
Proactive topics and continued conversationReply directly, discuss later, dismiss once, or stop future outreach on a topic; keep discussing a question without having to make a decision immediately
Action on suggestions and result follow-upCarry out work when you accept or existing authorization allows it, deliver artifacts, and update related plans, progress, and ongoing interests

For example, it might notice that your weekly commitments exceed your available time, compare two ways to adjust them, and ask about priorities. Thinking something through, planning ahead, or learning something important can be a complete contribution without turning the conversation into an execution task.

Stage 3: Continuous Self-Improvement

Let experience and feedback improve the next interaction and reduce repeated mistakes. Self-improvement always serves the purpose of being a better partner to you.

Planned featureWhat it will let you do
Automatic reflection on outcomes and feedbackIdentify useful lessons from failures, unsatisfactory results, and corrections without an extra request to reflect each time
Classification and application of experienceUpdate user understanding, participation preferences, or working methods in the appropriate place, and apply them in later relevant situations
Improvement history and resultsSee what changed, why, when it was applied, and what happened; clearly distinguish changes whose effects are still unverified
Correction, disabling, and reversal of experienceCorrect a mistaken lesson, disable an unsuitable strategy, or undo a change while preserving edits you made afterward

Stage 4: Easy to Start and Reliable Over Time

Planned featureWhat it will let you do
Partner setup and capability statusProvide necessary background, confirm a goal and participation scope, and see which capabilities your entry point supports, whether background work is running, and why it is paused
Recovery and upgradesKeep user understanding, plans, ongoing interests, and improvement history across restarts, connection failures, and upgrades; see failures and avoid duplicate notifications or actions on recovery

Persistence, stop controls, and permissions accompany each stage. Stage 4 checks their reliability during continued use. Support across runtimes and channels will follow actual validation; full 1.0 capabilities are not promised for every entry point at once.

How We Will Assess 1.0

Completing the features is followed by these core acceptance cases and continuous user trials. These are requirements, not results that have already passed.

Core caseRequired result
Understanding survives new conversations and changed plansRelevant background is used, corrections take effect, and expired or excluded facts no longer drive suggestions
Useful initiative without a new instructionIndependently analyze existing information and start a useful discussion, report relevant changes, and act when appropriate; stay quiet without new value
A real problem improves later behaviorA genuine correction or error leads to a traceable improvement that helps in another relevant situation and can be reversed
Reliable continuation after interruptionsStop requests, revoked permissions, restarts, and upgrades preserve correct state without false success or duplicate actions

After 1.0

Actual usage will guide broader data sources, services, runtimes, and device support, along with better judgment about initiative and improvement. Plugins, skills, SDKs, and the ecosystem will continue extending what your partner can do. Ordering and dates will be updated separately.

Released under the MIT License.