Scope: GoreeCloud AI product requirements and implementation maturity, November 11, 2026. This document describes the desired combined capabilities of a unified AI platform. It is not a claim that every item is built, tested, integrated, deployed, or supported on every operating system.
Inference foundation: Ollama, locally installed Qwen and Gemma models, and GoreeCloud Smart Model routing. Optional providers, specialized models, and integrations require separate installation, consent, licensing, and validation.
- Source implemented / CI tested: Function is represented in the merged repository and exercised by automated tests using a fake Ollama backend. Live Qwen/Gemma performance remains unverified.
- Planned: Product target; no verified implementation within this repository.
- Conditional/experimental: Dependent on platform, model, tools, explicit permissions, or unresolved security assessment. No deployment implied.
Source implemented / CI tested: Ollama model discovery; local chat completions and SSE streaming; installed-model validation; manual model selection; Qwen/Gemma Smart Model family routing; Fast/Balanced/Quality preference profiles derived from model-name size labels; routing explanations; inference-free route preview; local service health check.
Planned: Evaluated model benchmarks, context-aware routing, GPU/memory-aware scheduling, model version download UI, quantization-aware policies, model prewarming, resource concurrency budgets, provider-compatible adapters, cloud-provider permissions, reasoning-depth presets, multimodal model capability discovery, cost governance and fallback diagnostics.
Source implemented / CI tested: Persistent SQLite conversations; create/list/read/delete; first-message titles; contextual multi-turn requests; streaming Stop control; conversation search; single-conversation JSON export/import; atomic exchanges; optimistic concurrent-write conflict guard; bounded output storage; text-only safe rendering; keyboard message submission.
Planned: Temporary chats, conversation branching/forking, queued prompts, richer search, conversation folders, tags, pins, archive, message edit/regenerate, multi-model response comparison, conversation sharing, transcript references, document citation UI, multi-device synchronization and resumable streams.
Planned: Private and shared workspaces, project instructions and templates, workspace-specific model and tool policies, notes, explicit user memory, memory search/retention, personal settings, team sharing, scoped resource permissions, offline/project exports, isolated memory stores and portable project backups.
Source implemented / CI tested: Explicitly submitted local text documents, browser-assisted import of bounded .txt/.md/.csv/.json files as inert text, document listing/deletion, bounded keyword search/excerpts (including authenticated POST search), opt-in document context in chat, source ID/title metadata. An additional opt-in experimental Ollama embedding index and cosine semantic retrieval API now exists; it has mock-backed CI tests but no real-model quality verification. Keyword retrieval remains the default.
Planned: PDF/Word extraction, OCR, real-world semantic retrieval benchmarking, hybrid retrieval, BM25, reranking, source-grounded citations, retrieval evaluation, document sync with consent, metadata filtering, full-document context, classification/extraction and multi-document synthesis.
Planned: Custom assistants, agent templates, agent-scoped tools/knowledge, reusable skills, MCP tool discovery, planning, subagents, delegation, multi-agent coordination, visual flow builder, task checkpoints, human approval, agent execution budgets, cancellation, tool audit trails and sandboxed environments.
Conditional/experimental: Browser/computer-use agents, terminal actions, application automation and privileged tools require Wardveil review, isolation, explicit permissions and observable rollback.
Planned: MCP servers, OpenAPI/REST tools, authenticated connector management, per-user credentials, secret isolation, scope-limited tool permissions, OAuth integrations, webhook actions, external database connectors, Git/Drive/email/calendar/task services, auditable tool invocation and GoreeCloud Mesh protocols.
Planned: Search providers, live-source reading, structured webpage extraction, source provenance and citations, research plans, multimodal search, site-to-knowledge ingestion, source freshness and robots/consent policies.
Conditional/experimental: Authenticated scraping and browser automation require explicit opt-in and network/credential isolation.
Planned: Code explanation/review, repository context, debugging, test generation, editable artifacts, syntax-highlighting, diagram rendering, Docker or sandbox interpreters, file editing, preview servers and safe terminal integration. Local model outputs remain untrusted.
Planned: Image understanding, screenshot QA, OCR, image generation/editing through appropriate model/provider, speech-to-text, text-to-speech, live voice chat, transcription, meeting summaries, speaker labeling, video frame extraction and source-grounded media Q&A.
Conditional/experimental: Voice cloning, desktop screen capture, meeting recordings and video integrations require consent, supported operating systems, retention controls, and purpose-specific approval.
Planned: AI notes, task lists, recurring reminders, shared workspaces, calendar integrations, event/attendee support, project planning, communication channels, group chat with AI assistants, notifications, agent/task activity and protected sharing.
Planned: Cron/interval schedules, event/webhook triggers, durable jobs, run history, retries, idempotency, cancellation, task queue visibility, notifications, monitoring, and agent-run budgeting. Automation is never claimed active without a running scheduler and verified configuration.
Source implemented / CI tested: Local chat/model/conversation/document endpoints with optional shared API token and versioned JSON export/import.
Planned: Versioned public API contracts, SDKs, embeddable chat widgets, developer webhooks, role-scoped API credentials, extension catalogs and production-grade request limits.
Source implemented / CI tested: Optional shared API token for the local single-user preview. This is not per-user authentication.
Planned: GoreeCloud Identity SSO/MFA/passkeys, group permissions, user provisioning, resource ACLs, administrative settings, session management, access logs, API-key scopes, moderation and audit functions.
Source implemented / CI tested: Local storage by default, deliberate knowledge opt-in, local model routing, no model-weight downloads by the server, explicit import/export, plaintext-at-rest warnings.
Planned: Consent/retention enforcement, privacy dashboards, privacy-preserving logs, deletion/export verification, data minimization, remote-provider data sharing controls, cross-workspace isolation and personal-data governance. GoreeShield governs privacy, not cybersecurity.
Source implemented / CI tested: Local-only deployment guidance, safe DOM text rendering, bounded text inputs, conversation write integrity checks, stream failure handling and API token check.
Planned: Threat modeling, production IAM integration, abuse/rate limiting, CSRF/session hardening, prompt-injection resilience, network egress allowlists, secret vault integration, audit trails, malware/file validation, tool sandboxing, vulnerability scans and incident response. Wardveil governs cybersecurity and must enforce applicable GoreeShield privacy policies.
Planned: Verified service discovery, typed contracts, event routing, workflow coordination, synchronization and cross-product data exchange with explicit identity/privacy/security enforcement. Mere imports or naming do not establish Mesh integration.
Source implemented / CI tested: Functional dark responsive browser interface, accessible labels, conversation controls and knowledge dialog. This does not establish full GoreeCloud UI compliance.
Planned: Verified centralized design tokens, theme system, cross-platform components, accessibility audits, rich Markdown/code rendering, onboarding, premium visual QA, mobile/desktop layouts, high-contrast/reduced-motion states, keyboard navigation and genuine native-client experiences.
Source implemented / CI tested: Local offline evaluator for judged retrieval rankings, computing Hit@K, Precision@K, Recall@K, MRR@K, MAP@K and nDCG@K from bounded JSON fixtures. Its included fixture is synthetic and does not prove real-model quality.
Planned: Real Ollama retrieval corpus evaluation, token usage, latency/throughput measurements, real-model benchmarking, response quality evaluations, routing statistics, operational dashboards, metrics/logging/tracing, error analysis and privacy-limited diagnostics.
Source implemented / CI tested: Python packaging, container templates, local Docker Compose wiring, CI Python test matrix and syntax checks. Actual container execution and installed-model inference require separate real-host testing.
Planned: Verified Linux/Windows desktop packaging, Android/iOS native or individually qualified clients, backups/restores, migrations, signed releases, vulnerability/SBOM checks, health monitoring, GPU profiles, data encryption choices, upgrade/rollback and production readiness.
GoreeCloud AI is a hobby-owned, self-hosted, privacy-first product. Build third-party features through compatible licensed adapters or independently written code. Do not paste third-party source code without provenance/license review. Model weights and Ollama have independent obligations.
Always distinguish: vision → planned → implemented source → CI tested → real-runtime tested → released → deployed. The first four GoreeCloud pillars remain distinct, with Identity included in the mandatory integrated systems.
See Implemented Features, Planned Features, Architecture, Security, Smart Model, Conversation Portability, and Offline Retrieval Evaluation.
Priority engineering milestones: (1) real-model/local-host tests and recovery, (2) GoreeCloud Identity/GoreeShield/Wardveil integration gates, (3) evaluated local semantic RAG, (4) scoped workspaces and agent tool permissions, (5) native clients and production acceptance checks.