A Code Assistant

AI-powered developer workspace for GitHub repos, project intelligence, code review, and team collaboration.

Build faster with AI that understands your real project context — not just generic chat. Connect your GitHub repositories, analyze codebases, review pull requests, and collaborate inside one workspace designed for real engineering work.

Why this project matters

Most AI tools stop at chat. Real development happens across repositories, files, pull requests, issues, and team workflows. AI Code Assistant brings project-aware AI into the full software lifecycle so developers can explore code, understand architecture, review quality, and collaborate with context.

What it does

• Chat with AI over project and repository context
• Analyze GitHub repositories, issues, and PRs
• Import projects and search across imported files
• Run AI code review and quality analysis
• Collaborate in workspaces with permissions and activity history
• Extend the platform with plugins and capability-based integrations
• Support Stellar / Soroban-aware development workflows

Try it in 5 minutes

# 1. Clone the project
git clone https://github.com/AyindxLab/ai-code-assistant.git
cd ai-code-assistant

# 2. Create a virtual environment
python -m venv .venv
source .venv/bin/activate

# 3. Install dependencies
pip install -r requirements-dev.txt

# 4. Configure environment
cp .env.example .env

# 5. Run the app
python run.py

Open http://localhost:5000 in your browser.

Who it's for

• Developers who want AI help grounded in their actual codebase
• Teams reviewing Pull Requests and project quality
• Engineers working across GitHub repositories and project files
• Builders experimenting with AI workflows, plugins, and repo intelligence

Why it stands out

• Project-aware AI instead of generic chat-only tools
• GitHub repo and PR intelligence built into the workflow
• Structured findings with review history and quality signals
• Team collaboration and workspace-level permissions
• Extensible plugin architecture for future integrations

Marketplace and plugin configuration

Plugin installation and outbound requests are opt-in. The defaults are fail-closed, so a fresh install cannot reach a marketplace or the network:

Setting	Default	Why it is safe
PLUGIN_CATALOG_FILE	empty	Remote/catalog installation is disabled until an operator points at a vetted catalog.
PLUGIN_MARKETPLACE_ENABLED	1	Read-only discovery is available, but the shipped catalog is empty.
PLUGIN_MARKETPLACE_CATALOG	plugins/marketplace.json	Offline catalog file; entries must validate against plugins/plugin.schema.json.
PLUGIN_MARKETPLACE_URL	empty	Remote marketplace fetching is disabled; when set it must be HTTPS, allowlisted, timeout/size-bounded, and redirect-free.
PLUGIN_DISCOVERY_DIR	plugins	Read-only directory scanned for manifests by GET /plugins/discover; nothing is installed or executed.
PLUGIN_TRUST_POLICY	if-present	Accepts unsigned local plugins, but always refuses forged or untrusted signatures. Use required in production.
PLUGIN_TRUSTED_KEYS	{}	No publisher is trusted by default; keys are validated at startup.
PLUGIN_NETWORK_ALLOWLIST	empty	Every plugin outbound request is denied. Add only the hosts a plugin needs.
PLUGIN_NETWORK_HTTPS_ONLY	1	Plugin traffic must use HTTPS.
PLUGIN_NETWORK_ALLOW_PRIVATE	0	Blocks private, loopback, and link-local targets.
PLUGIN_EXECUTION_TIMEOUT	5	Abandons a plugin callback after five seconds so it cannot stall a request.
PLUGIN_CONFIG_MAX_BYTES	65536	Caps the serialized size of a workspace plugin configuration.
PLUGIN_CONFIG_MAX_KEYS	200	Caps the number of top-level configuration keys.
PLUGIN_CONFIG_MAX_DEPTH	8	Caps configuration nesting depth.
PLUGIN_ERROR_REPORT_LIMIT	100	Caps owner-visible error reports (hard ceiling 500).

The catalog is read from disk for each install request; there is no active cache TTL setting yet (caching is tracked in #377). See the plugin guide for the full reference, including signature and catalog formats.

Observability and billing

Every LLM-backed path appends a metadata-only row to an append-only usage ledger; prompts and completions are never stored. Dashboards, budgets, exports, and cost estimates are computed from that ledger. Cost figures are estimates from a documented price table, not provider invoices.

Setting	Default	Purpose
METRICS_ENABLED	1	Serves protected Prometheus metrics at /metrics; 0 returns 404.
METRICS_TOKEN	empty	Bearer token for scrapers. Empty falls back to admin-session auth.
USAGE_RETENTION_DAYS	365	Minimum age before flask usage purge may delete ledger rows.
BUDGET_ENFORCEMENT_ENABLED	1	Refuses LLM calls that would exceed a monthly token/cost budget (fail closed).
BUDGET_DEFAULT_WARN_THRESHOLD	0.8	Default warning fraction for new budgets.
BILLING_IMPORT_MAX_BYTES	1048576	Maximum provider billing CSV upload size.
BILLING_IMPORT_MAX_ROWS	5000	Maximum billing CSV rows.
CURRENCY_CODE	USD	Currency code shown on every cost figure.

See Observability for the ledger, metrics, dashboards, and export contract, and Billing for the price table, budgets, and reconciliation.

Documentation

• Chat feature guide
• GitHub integration
• Stellar / Soroban tooling
• Plugin system
• Architecture
• Security model
• Soroban workflow guide
• Observability design
• Billing
• Cost precision
• Usage and billing data privacy
• Roadmap

License

MIT