ai-code-assistant

AI Code Assistant is a developer-focused AI code intelligence platform that provides AI-powered coding assistance, repository analysis, code review, and team collaboration. It is built incrementally across eight phases using Python 3.12, Flask 3, PostgreSQL, and a clean vanilla CSS/JS frontend with SSE streaming. Core capabilities include a per-user chat interface with live streaming responses, a prompt library, AI code generation and analysis tools (explain, refactor, find bugs, optimize, add comments, write documentation, draft commit messages), and file upload analysis. GitHub integration supports OAuth connection, repository browsing, commit history, issues, and pull requests with AI-powered analysis. PR reviews return structured findings labeled [CONFIRMED] versus [SUGGESTION], with bounded context to avoid uploading entire repositories. Access tokens are encrypted at rest with Fernet. AI workspaces allow project import from GitHub or archive upload, lazy file exploration, project-wide search, dependency inventory, AI project chat, and a health dashboard. Imported content is explicitly framed as untrusted data to resist prompt injection. Team collaboration (Phase 7) provides member lifecycle management, email invitations with hashed tokens, role-based permissions, notifications, @mentions, threaded project discussions, and an append-only activity audit log. The plugin system (Phase 8, in progress) features validated manifest support, a registry, a capability model with explicit grants, and an event dispatcher. Plugins can subscribe to events such as project.created, review.completed, and stellar.analysis.completed. Lifecycle hooks, per-workspace configuration, PEP 440 compatibility enforcement, and owner-scoped audit trails are supported. Stellar/Soroban developer tooling includes read-only Horizon and Stellar RPC services, contract and account inspection, XDR decoding for transaction envelopes and LedgerKey/SCVal data, an offline mock network, heuristic project detection for Soroban contracts, and Stellar-aware AI analysis. Mainnet is never an implicit default and raw URLs are rejected. Infrastructure uses multi-stage Docker deployment, GitHub Actions CI with ruff linting, black formatting, and pytest, and a non-root runtime user.