AI Tools & Providers
Company-approved AI coding tools — IDEs, CLIs, and the model providers each one supports. Includes access instructions, privacy requirements, and billing policies.
This page covers the AI-assisted coding tools approved for use at Aircury, how to get access to each one, and how to connect them to the model providers we use. All tools on this list are pre-approved — if you want to use something not listed here, ask Jose first.
Privacy First
Client code is sensitive. Before using any AI tool on a client repository, make sure you have enabled the privacy settings described below. When in doubt, ask before you paste.
IDEs
Cursor
Cursor is an AI-native code editor built on VS Code. It is the primary IDE at Aircury.
Getting access
Each employee can have a Cursor account tied to their corporate email. Ask Jose to assign you a licence on the company account. Do not use your personal Cursor account on client repositories.
Required privacy settings
Configure these before working on any client code:
Enable Privacy Mode
Cursor Settings → Privacy → Privacy Mode → On
This enables zero data retention with Cursor’s model providers. Without it, your code and prompts may be used to improve Cursor’s models.
Disable commit and PR attribution
Cursor Settings → Agents → Commit Attribution → Off
Cursor Settings → Agents → PR Attribution → Off
This prevents Cursor from adding attributions to your commits and pull requests.
Review other settings periodically
Cursor adds new settings as the product evolves. Periodically review the Agents and Privacy sections for anything that should be disabled under our policy, and flag it to Jose if you find something.
Model providers
Cursor’s Pro licence includes credits that cover access to models from multiple providers with no extra configuration:
| Provider | Models included |
|---|---|
| Anthropic | Claude Opus 4.6, Claude Sonnet 4.6 |
| OpenAI | GPT-5.4, GPT-4o mini |
| Gemini 2.5 Pro, Gemini 2.5 Flash | |
| xAI | Grok Code, Grok 3 Mini Beta |
| DeepSeek | DeepSeek v3 |
You can also bring your own API key (Settings → Models → API Keys) for OpenAI, Anthropic, Google, or Azure OpenAI endpoints. Custom OpenAI-compatible endpoints (e.g. Ollama, OpenRouter) are supported too.
Learning resources
There are recorded walkthrough videos in the Seminario IA folder — check those for a hands-on introduction to Cursor’s features.
Antigravity
Antigravity is Google’s agent-first AI IDE — a VS Code fork with deep multi-agent capabilities, released in November 2025 as a direct competitor to Cursor. It is available free during public preview and is pre-approved at Aircury.
Getting access
Antigravity is free to download and use. Sign in with your corporate Google/Workspace account (@aircury.com or @elemwave.com) — this is required to get enterprise data protections (see Privacy below). Do not use a personal Google account on client code.
Features
- Up to 5 agents running simultaneously across different workspaces
- Built-in browser for agents to test UI and run integration checks
- Two modes: Planning (design and architecture) and Fast (execution)
- MCP server support via a growing ecosystem of 1,500+ integrations
Model providers
| Provider | Models |
|---|---|
| Gemini 3.1 Pro, Gemini 3 Flash | |
| Anthropic | Claude Sonnet 4.6, Claude Opus 4.6 |
Privacy
When signed in with a Workspace or GCP account, your interactions are not used to train models and Google employees cannot access your data. This matches the enterprise data protections available on other Google products.
Personal accounts are not safe for client code
If you sign in with a personal Google account, your interactions may be used for model training. Always use your @aircury.com or @elemwave.com account in Antigravity.
CLIs
Command-line AI coding tools let you work with AI agents directly from your terminal, inside any editor or workflow.
Claude Code
Claude Code is Anthropic’s official agentic CLI. It is pre-approved at Aircury and is the recommended CLI for day-to-day AI-assisted development.
Getting access
Claude Code requires a Claude Pro subscription (minimum $20/month) or direct API credits. Follow the company reimbursement process described below if you need your own account.
Supported providers
Claude Code is natively locked to Anthropic models. It supports three access paths:
| Path | How to configure |
|---|---|
| Anthropic API (direct) | Set ANTHROPIC_API_KEY in your environment |
| Amazon Bedrock | Set CLAUDE_CODE_USE_BEDROCK=1 + AWS credentials |
| Google Vertex AI | Set CLAUDE_CODE_USE_VERTEX=1 + GCP project credentials |
All three paths use Claude models only (Sonnet 4.6, Opus 4.6, Haiku 4.5). Routing to non-Anthropic models is possible via AI gateways (LiteLLM, OpenRouter, Cloudflare AI Gateway) but is not a standard supported workflow.
Privacy
Anthropic does not use API inputs or outputs to train models by default. Data is retained for up to 30 days for abuse monitoring. Bedrock and Vertex usage follows AWS and GCP data policies respectively, which generally provide stronger enterprise guarantees.
Codex CLI
Codex CLI is OpenAI’s official terminal-based coding agent. At Aircury we primarily use it via the company’s shared ChatGPT Plus account.
Getting access
The company has a shared ChatGPT Plus account — credentials are in Vault. If your work depends heavily on Codex, ask Jose for a dedicated account (subject to the reimbursement policy below).
Supported providers
| Provider | How to connect |
|---|---|
| OpenAI API | Set OPENAI_API_KEY or log in with your ChatGPT account |
| Local models | Via Ollama or LM Studio with --oss flag |
| Custom endpoints | Any OpenAI-compatible endpoint via ~/.codex/config.toml |
The default model is codex-mini-latest. GPT-5 family models are available on Plus and Pro plans.
Privacy
- Free / Plus plans: Data may be used to train OpenAI models unless you opt out in account settings.
- API usage: Data is not used for training by default; retained up to 30 days.
- Business / Enterprise plans: Data is contractually excluded from training.
ChatGPT privacy settings
If you are using a personal or company ChatGPT account, verify that “Improve the model for everyone” is disabled in Settings → Data Controls to comply with company policy.
OpenCode
OpenCode is an open-source, terminal-based AI coding agent that supports over 75 model providers. It has no vendor lock-in and is pre-approved at Aircury.
Getting access
OpenCode is free to install. You connect it to providers by running /connect inside the TUI and selecting your provider. No single subscription is required — it works with whatever keys or plans you already have.
Supported providers and connection options
OpenCode supports any of the following:
| Option | What you get | Notes |
|---|---|---|
| Bring Your Own Key | Full access to any of 75+ providers (OpenAI, Anthropic, Google, AWS Bedrock, Groq, DeepSeek, Ollama, etc.) | You pay per token at each provider’s rates |
| OpenCode Zen | Curated set of pre-tested models from multiple providers, pay-as-you-go | Zero-retention policy; models hosted in the US; some free-tier models have data training caveats (see below) |
| OpenCode Go | GLM, Kimi K2.5, MiniMax M2.5, MiniMax M2.7 | $10/month subscription (first month $5); models hosted in US, EU, Singapore; designed for cost-effective access to open-source models |
OpenCode Go details
OpenCode Go ($10/month after the first $5 month, currently in beta) gives you access to:
- GLM-5 — Zhipu AI’s flagship model
- Kimi K2.5 — Moonshot AI’s coding-focused model
- MiniMax M2.5 / M2.7 — MiniMax’s latest generation
Usage is rate-limited to $12 per 5 hours, $30/week, and $60/month of token-equivalent usage.
Privacy — free models
Several models available free through OpenCode Zen explicitly allow data collected during their free period to be used for model training. This includes:
- GLM 4.7 Free
- Kimi K2.5 Free
- MiniMax M2.1 Free / M2.5 Free
Do not use free-tier models on client code
The free versions of GLM, Kimi, and MiniMax allow training data use. Never use these free-tier models with client repositories or any proprietary code. Use the paid OpenCode Go subscription or your own API keys instead.
Providers
This section covers AI model providers that are available company-wide and that require individual setup. For provider-specific pricing details, see AI Pricing Models.
AWS Bedrock
AWS Bedrock is the recommended provider for work that requires the strongest privacy guarantees. It gives you access to Claude (and other) models through AWS infrastructure with contractual data isolation — Anthropic itself cannot see your prompts or completions.
Why Bedrock for privacy
Bedrock’s privacy model is structurally stronger than using the Anthropic API directly:
| Guarantee | Direct Anthropic API | AWS Bedrock |
|---|---|---|
| Data not used for training | Yes (by default) | Yes (guaranteed) |
| Provider cannot see your prompts | No — Anthropic processes them | Yes — model runs in isolated AWS-owned accounts; Anthropic has no access |
| Data stays in your region | Depends on plan | Configurable per region |
| Enterprise compliance (SOC 2, ISO 27001, HIPAA) | Enterprise plan only | Included by default |
| Customer-managed encryption keys | No | Yes (via KMS) |
Available models
| Model | Input / 1M tokens | Output / 1M tokens |
|---|---|---|
| Claude Opus 4.6 | $5.00 | $25.00 |
| Claude Sonnet 4.6 | $3.00 | $15.00 |
| Claude Haiku 4.5 | $1.00 | $5.00 |
Bedrock is pay-per-token, not subscription
Unlike Cursor Pro or Claude Pro (fixed monthly fee with usage quotas), Bedrock charges per million tokens consumed — every request costs real money. It can become significantly more expensive than a subscription for heavy day-to-day use. Use it when privacy requirements justify the cost, not as a default. See AI Pricing Models for a full comparison.
Getting API credentials
Request an API key
Ask Alejandro Rey, Raul, or Pedro to generate a Bedrock API key for you.
Log it in the tracking spreadsheet
Add your name and key ID to the Bedrock API Keys spreadsheet. This lets the team audit active keys and rotate them when needed.
Configure your tool
Add the key to OpenCode via /connect → AWS Bedrock, or follow the setup instructions for whichever tool you are using.
Billing & Reimbursement
The company reimburses personal subscriptions paid under the company name and corporate email at the end of the month via payroll.
| Rule | Detail |
|---|---|
| Monthly limit | £30€/month per person |
| Billing name | Must be under the company name and your company email |
| No annual plans | Do not sign up for yearly subscriptions (You might want to switch to another provider) |
| Claim form | Submit your invoice at forms.gle/CTLppwJ6UZMUv1N49 |
| Over limit? | Ask Jose before committing to anything above €30/month |
Pre-approved tools (no prior approval needed):
- Cursor
- Claude Code
- ChatGPT (Codex)
- OpenCode
- Antigravity
For any tool not on this list, ask Jose — new tools are regularly added.
Why self-pay?
This policy exists because centrally managing corporate accounts has become increasingly difficult as vendors implement complex billing and seat-management flows. Self-pay with reimbursement gives each employee flexibility while keeping the company in control of what’s approved.