Models
The models Meevra items are written for, where they run and how to prompt them.
Anthropic
Default model on Pro, Max, Team, Enterprise and the Anthropic API; default effort medium.
Not a default on any plan; select with /model fable. Usage may bill to usage credits.
Select with /model sonnet; default effort medium in Claude Code.
OpenAI
OpenAI recommends it for complex coding and agentic work when available to the account; effort Light to Ultra (Ultra splits work across subagents).
Included for Plus in Work and Codex; needs Codex CLI 0.153.0 or newer.
For focused, repeatable tasks; effort up to Max (no Ultra).
Anysphere
Cursor's own model, drawn from the plan's usage pool. Cursor marks no default model; Claude Opus 5.5, Sonnet 5.5, Fable 5.1, Gemini 4 Argon, Gemini 3.8 Flash and GPT-5.6 models are also listed (no GPT-6 model listed on 2026-10-15).
Available on all Antigravity plans.
Google AI Ultra and Enterprise only; not in the Antigravity IDE.
With Claude Sonnet 5.5, on Google AI Pro (non-trial), Ultra and Enterprise.
Lovable
Lovable does not publish which model runs its agent, so content targets the product, not a model.
StackBlitz
Balanced for everyday building; Bolt picks the underlying model.
Maximum reasoning for large apps, complex features, refactoring and open-ended tasks.
Moves a prompt, skill or API call to DeepSeek, Qwen, GLM or Kimi with the right model, effort and parameters, then checks three real inputs side by side before you switch.
Fill-in instructions for a ChatGPT Project that keep the assistant on one job, in one answer shape and grounded in your project files, plus the steps to turn the same text into a skill you can share.
A fill-in CLAUDE.md that tells a coding agent your stack, your exact commands, what needs your approval and how to report each task, so it stops adding packages and reporting passes it never saw.
Paste a Custom GPT's configuration and get its replacement: cleaned Project instructions, a plugin-ready SKILL.md, a knowledge-file map, a parity test and a cutover checklist.
Answer five questions in chat and get one AGENTS.md that every coding agent reads, plus thin CLAUDE.md and Cursor rule files only where you need them, with every rule checkable against a diff.
One AGENTS.md that every coding agent reads, plus your repeated procedures turned into starter skills in the folders each agent loads.
Turns a problem and its limits into an algorithm that provably fits them, with an invariant, a hand trace and a brute-force comparison test.
A two-page onboarding guide to an unfamiliar repo that separates what the code shows from what is guessed, ending with a safe first change.
Converts saved Custom GPT configurations into ChatGPT Projects or plugin-ready skills, with a parity test against the old GPT and a cutover checklist.
From hand-merged exports to a re-runnable pipeline whose report totals tie back to the source files, with checks that stop bad data.
Explore first, pin current behaviour with tests, then make the smallest reviewed change in code nobody fully understands.
A merge verdict on a diff, judged against your repo's own rules, with a trigger and a fix for every blocker and a list of what the diff is missing.
AGENTS.md, CLAUDE.md and Cursor rules written from the code that is actually in your repo, with evidence for every convention and commands that actually run.
A refactor plan made of small steps that each keep the system working, with the behaviour contract and the first tests to write.
Runs a scoped refactor in your coding agent: every instance counted first, one shown for approval, no drive-by edits, real test output at the end.
A disciplined search for the real cause of a bug, with the cheapest test for each suspect and no fix until the cause fits in one sentence.
Turns an idea or a repeated chat routine into one SKILL.md that triggers on the right requests, with a test set you run before sharing it.
Bundles finished skills into a Claude Code or ChatGPT/Codex plugin and audits collisions, conflicts, secrets and installs before anyone else uses it.
A feature idea becomes a spec with testable acceptance criteria, a data model and an API contract, and the brief's contradictions spelled out.
Your acceptance criteria become failing tests you approve first; the agent then implements until they pass, with no silent test edits.
Write the short standing brief that Lovable or Bolt reads with every prompt, so the builder keeps your rules and stops repeating the same mistakes.
Turn a rough app idea into one spec that Lovable, Bolt or Cursor can build from without guessing: every stated need kept, every rule numbered, every record's lifecycle defined, every gap decided and marked.
Turn a messy list of changes for a Lovable or Bolt app into small prompts in a safe order, each naming what to change, what must stay and how to check it.
Writes Bolt prompts that connect AI, email, maps or any other non-payment API with the key on the server, cost caps, failure handling and a test prompt.
An evidence-based audit of an AI-built app: ranked issues with reproduction steps, fix prompts with acceptance checks, and a ship or no-ship verdict.
Designs a Lovable app's tables, constraints and row level security rules before the pages exist, with a migration template and a two-user test plan.
Writes Lovable prompts that replace default styling with one central design system, refine each page with every state designed, and keep new pages consistent.
Turns a Bolt bug into a scoped debug prompt with a likely cause to confirm first, plus the follow-up prompts for every outcome.
Plans a new Bolt build as a stack lock plus an ordered prompt sequence with a check after every step and a coherence audit every three prompts.
Turns the pre-launch checklist into Plan-mode audits, Build-mode fixes and checks you do yourself, so a working Lovable demo becomes an app real users can rely on.
Sets up routing, access, state and layout for a 5 to 15 page Bolt app, then builds every page from one reference page.
Turns an app idea into an ordered set of Lovable prompts for a weekend MVP, each with a preview check, with every requirement kept, permission and limit checks, and a launch block.
Writes the Bolt prompts and the go-live test matrix for Stripe or Paddle payments where only a verified webhook grants access and refunds take it away.
Maps what survives, changes and goes in an existing Lovable app, then writes the prompt sequence that turns it into the new product without losing users or data.