How much does Vyre IDE cost?

Vyre IDE has a free tier for individuals and teams. Paid plans add higher usage limits, premium model capacity, team administration, and priority support.

Do I need to pay for AI usage?

AI usage depends on the plan and model path you choose. Vyre also supports bring-your-own-key workflows on paid plans for teams that want to use their own provider accounts.

Private project files are not used for Vyre product training. Selected context and prompts may be sent to your configured model provider, and ignored files should stay excluded.

Can I use Vyre IDE offline?

Core desktop editing can work locally, but collaboration, account, and hosted AI features require internet access. Local model workflows can use providers such as Ollama when configured.

How do I migrate from another IDE?

Use the migration guides for Cursor, Windsurf, Zed, JetBrains, Kiro, Roo Code, and related tools. Most pilots should start with one real project before a team-wide rollout.

Cloud Agent Best Practices

Cloud agents are powerful but require thoughtful task design. Follow these practices to get reliable, cost-effective results.

Designing good tasks

The most important factor in cloud agent success is task quality. Cloud agents don't have you sitting next to them to clarify — they need a complete, unambiguous task description.

Be specific

'Fix the auth bug' is vague. 'Fix the bug in AuthService.refreshToken() where sessions expire 1 hour early' is actionable.

Include success criteria

'...and all existing tests should pass' gives the agent a clear way to verify its own work.

Set scope boundaries

'Only touch files in src/api/. Do not modify the frontend.' prevents agents from going too far.

Specify the output

'Open a PR with the changes' or 'commit to the feature/auth branch' tells the agent what "done" looks like.

Good task template

code

[Context]: [1-2 sentences about the codebase context]
[Task]: [What specifically needs to be done]
[Constraints]: [What not to change, edge cases to handle]
[Verification]: [How to verify the work is correct]
[Output]: [What to do when done — commit, PR, issue]

Example:

code

Context: We have an Express.js REST API with JWT authentication.
Task: Add rate limiting to all /api/auth/* routes. 
      Limit to 5 requests per minute per IP.
Constraints: Use the existing `express-rate-limit` package 
             (already installed). Don't touch any other routes.
Verification: Run `npm test` — all auth tests must pass.
             Manually verify by hitting /api/auth/login 6 times quickly.
Output: Open a PR titled "feat: rate limit auth routes"

Cost optimization

Choose the right model

Use faster, cheaper models for simpler tasks:

Claude Haiku — formatting, renaming, simple edits, documentation
Claude Sonnet — most coding tasks (default)
Claude Opus — only for the most complex architectural decisions

Specify the model per task:

bash

vyre agent run --task "Format all files" --model claude-haiku-3-5

Set max duration

Always set --max-duration to prevent runaway tasks:

bash

vyre agent run --task "..." --max-duration 300  # 5 minutes

Tasks that exceed the limit are automatically stopped. You keep a transcript of what was done so far.

Use no-write for research tasks

If you want the agent to analyze code and report findings without making changes, add --no-write:

bash

vyre agent run --task "Audit the codebase for security issues and report findings" --no-write

Read-only tasks use less tokens because they don't need to plan and verify writes.

Reliability patterns

Always include a test verification step — agents that can run tests catch their own mistakes.
Break large tasks into smaller runs — a 60-minute task is harder to recover from than three 20-minute tasks.
Use branches, not main — always have cloud agents work on a feature branch and merge via PR review.
Monitor the first few runs — watch new automations closely before letting them run unsupervised.

The best cloud agent runs have clear task descriptions, a verification step (running tests), and a defined output (a PR). These three elements make agent runs highly reliable.