Views
No views yet
agents/manifest.json and the universal harness. Reasoners read, builders propose, verifiers inspect, operators execute approved allowlisted actions, and archivists record local metadata. No agent receives personal credentials, arbitrary network access, arbitrary shell access, or permission to export private source by default.models/PACKS.md. The owner's unfinished Liquid model is deliberately excluded.1npm install
2npm run doctor
3npm starthttp://127.0.0.1:4173/. The doctor reports missing optional runtimes as warnings instead of hiding them. Import a model pack from the Model Lanes panel and use START / TEST RUNTIME.runtime/README.md and models/manifest.json for the adapter contract and model configuration. AIDE does not apply model-generated patches automatically.daemon/server.mjs is the first runnable native boundary for workspace and Git operations. It binds to loopback and exposes only bounded read-only endpoints. desktop/README.md defines the Tauri shell plan and acceptance gates; no compiled desktop binary is claimed in this pre-production release.harness/ contains the model-independent closed loop: guard, retrieve, plan, propose, verify, revise, test, review, and learn. Liquid is assigned to reasoning and verification; Qwen Coder is assigned to patch generation. The harness does not mutate model weights and never lets a model approve or apply its own changes.harness/credo.md; its research and copyright boundary is in harness/credo-research.md. It is mandatory harness context for every model role, alongside the role-specific SOP cards in harness/sops.json. harness/veritas.mjs uses calibrated task thresholds: 90% for ordinary explanation/code-change evidence and 98% for security, publishing, payment, and identity operations. These are evidence gates, not promises of universal model accuracy; failed gates produce abstention.npm run veritas executes the real local gate before a release or verified answer: compile checks, tests, Git whitespace checks, manifest validation, path boundaries, and secret scanning. It is intentionally allowlisted and does not execute arbitrary model-generated commands.community/journal-desk.md and community/casefile-schema.json.languages/manifest.json; models remain responsible for proposals, not syntax or semantic truth.debuggers/manifest.json and debugpy; breakpoints, stack inspection, and execution remain outside the model harness.training/manifest.json and training/README.md; it is an approval-gated, resume-safe control room for data, tokenizer, training, evaluation, and release jobs. benchmarks/ contains the first common smoke suite for all three public model packs. desktop/tauri.conf.json is the reproducible desktop-shell configuration; native compilation remains a platform gate until Rust/Tauri is installed.capsules/ adds reproducible offline workspace capsules: portable metadata for the exact model, runtime, Git revision, evidence hashes, tools, and Veritas result without exporting private source by default.benchmarks/live-results-2026-08-10.json: all three models passed function and planning smoke tasks, while all failed strict unified-diff formatting. AIDE therefore keeps every raw patch behind Veritas and human approval.ready with a local runtime endpoint.release/desktop-build-status.json. Installer bundles still require a desktop host with a graphical session.