Repository navigation
Releases
GitLearnOS v2 Developer Preview
GitLearnOS v2.0.0-dev.1
GitLearnOS v2 Developer Preview is the first GitHub release checkpoint for the v2 protocol.
GitLearnOS gives one capable, replaceable main AI agent a learner-owned Git memory. Learning can happen in class, on paper, in books, on practice platforms, in projects, with teachers, peers, or other AI systems; GitLearnOS keeps only the evidence that is useful for the learner's next decision and preserves it as inspectable, reversible state.
Downloads
Choose the package that matches what you want to do:
GitLearnOS-Starter-Template-2.0.0-dev.1.zip: recommended for most learners and AI-agent setups. It contains the v2 protocol, Quickstart, setup templates, the complete GitLearnOS Skill, bilingual documentation, and the generic agent/Git/search/automation adapters needed to deploy GitLearnOS into a learner-owned repository.GitLearnOS-DeepSeek-Harness-Plugin-2.0.0-dev.1.zip: the native DeepSeek Harness Developer Preview bundle. Extract it and install from its root withdsh plugin --profile web add ., then verify discovery withdsh --profile web --dump-config.
The automatically generated Source code (zip) and Source code (tar.gz) remain full repository snapshots for developers. They are not the recommended end-user packages.
Highlights
- Git-native v2 protocol:
GITLEARNOS.mdis the canonical behavior contract, with local or hosted Git as the core and GitHub treated as an optional adapter. - One-agent learning loop: organize evidence, diagnose the current gap, ask a targeted question, preserve answers or external feedback, recheck independently, and write back with one reversible Git commit.
- Learner-controlled writes:
safe-auto,preview, andmanualmodes keep automation bounded and visible. - Structured learning state: subject-specific goals, sources, knowledge, models, knowledge gaps, reviews, and useful learning events live under
subjects/<subject>/. - Differential diagnosis before durable gaps: competing explanations are tested before the system writes a knowledge-gap conclusion.
- Native DeepSeek Harness Developer Preview: an installable GitLearnOS bundle provides bounded observations, an atomic
learning_applytransaction, and an agent-maintainedNext uppanel. - RAG is a separate retrieval layer: Git remains the readable source of learning truth. RAG-Anything is the first explicitly supported option, not a lock-in.
- Verified background work: reminders and requested schedules are not treated as completed automation unless a real scheduler has actually run them.
- Migration path from v1: v2 can be adopted gradually without destructively rewriting existing learner evidence.
Developer Preview boundary
This is a pre-release. The Git learning loop is usable, but a complete deployment requires a verified local or compatible private RAG layer. The bundled DeepSeek Harness integration is also explicitly a Developer Preview. Multimodal input depends on the configured model or an authorized vision/OCR bridge.
GitLearnOS does not claim mastery from an immediate correct answer, does not fabricate scheduler execution or RAG access, and does not treat dashboards as a second source of truth.
Start here
QUICKSTART.md: canonical setup pathGITLEARNOS.md: v2 protocolMIGRATION-v2.md: migration from earlier layoutsLIVE-DEMO.md: end-to-end learning loop exampleadapters/deepseek-harness/README.md: Harness installation, limits, and verificationevals/README.md: documented scenarios
Official website: https://guojiz.github.io/gitlearnos/
Release target
This release is cut from the main branch after the September 2026 v2 protocol, Harness, RAG-verification, and Git/RAG separation work.