ScopeGuard
A local-first open-source AI tool for automatically detecting scope creep in freelance projects. It semantically compares client requests with contract terms and generates boundary responses and change-order quotations.
Timeline (last 90 days)
ScopeGuard is a 100% locally running, privacy-first open-source AI application designed for freelancers, design agencies and independent consultants to automatically detect scope creep.
ScopeGuard's core logic is to semantically compare new client requests against the signed project agreement or SOW, flag out-of-scope content and cite the specific clauses.
ScopeGuard is open-sourced under the MIT license, with a tech stack of Django 5 + React 19 + TypeScript, SQLite on the backend and a dark theme on the frontend.
ScopeGuard uses Ollama to run open-weight models locally (such as llama3:8b, qwen3:8b, gemma2:9b), and all inference requests stay on the local machine.
ScopeGuard's AI pipeline includes Ollama local LLM inference, ChromaDB vector retrieval, SentenceTransformers local embeddings, a deterministic NLP fallback engine and a deterministic pricing engine.
ScopeGuard uses the all-MiniLM-L6-v2 model for local embeddings; the model has about 22 million parameters and can encode in milliseconds on a CPU.
When Ollama is offline or the device battery is low, ScopeGuard automatically falls back to an internal regex and semantic rule engine, retaining 100% functionality while offline.
ScopeGuard's financial calculations use a deterministic pricing engine to prevent the LLM from hallucinating on financial agreements.
ScopeGuard's backend exposes endpoints such as /api/projects, /agreements, /analyses and /quotations through Django REST Framework, and PDF generation uses ReportLab.
In one user test, ScopeGuard flagged a client's new request as OUT_OF_SCOPE with 92% confidence, cited the exclusion clause and generated a change order quote of $1,350 (18 hours × $75/hour).
1 more timeline events
All Facts (11)
ScopeGuard's local LLM operation has certain hardware requirements, and the semantic judgment accuracy in fallback mode is somewhat reduced.
50%UnverifiedScopeGuard's core logic is to semantically compare new client requests against the signed project agreement or SOW, flag out-of-scope content and cite the specific clauses.
50%UnverifiedScopeGuard is open-sourced under the MIT license, with a tech stack of Django 5 + React 19 + TypeScript, SQLite on the backend and a dark theme on the frontend.
50%UnverifiedScopeGuard uses Ollama to run open-weight models locally (such as llama3:8b, qwen3:8b, gemma2:9b), and all inference requests stay on the local machine.
50%UnverifiedScopeGuard's AI pipeline includes Ollama local LLM inference, ChromaDB vector retrieval, SentenceTransformers local embeddings, a deterministic NLP fallback engine and a deterministic pricing engine.
50%UnverifiedScopeGuard uses the all-MiniLM-L6-v2 model for local embeddings; the model has about 22 million parameters and can encode in milliseconds on a CPU.
50%UnverifiedWhen Ollama is offline or the device battery is low, ScopeGuard automatically falls back to an internal regex and semantic rule engine, retaining 100% functionality while offline.
50%UnverifiedScopeGuard's financial calculations use a deterministic pricing engine to prevent the LLM from hallucinating on financial agreements.
50%UnverifiedScopeGuard's backend exposes endpoints such as /api/projects, /agreements, /analyses and /quotations through Django REST Framework, and PDF generation uses ReportLab.
50%UnverifiedScopeGuard is a 100% locally running, privacy-first open-source AI application designed for freelancers, design agencies and independent consultants to automatically detect scope creep.
50%UnverifiedIn one user test, ScopeGuard flagged a client's new request as OUT_OF_SCOPE with 92% confidence, cited the exclusion clause and generated a change order quote of $1,350 (18 hours × $75/hour).
50%