Plurai Simulation
Plurai Simulation tests AI agents in a simulated world of mocked tools and synthetic users, with plans from free to $210 a month.
What it does
Plurai Simulation builds a simulated world around an AI agent from the agent's own code: a digital twin of the agent, mocked tools, services and databases, and synthetic users with realistic goals. The agent runs its real tasks in an isolated sandbox, and each failure comes back with the code location that caused it and a fix checked against the full regression set.
The product ships as a plugin for Claude Code, Codex and GitHub Copilot, with Cursor listed as coming soon. One command maps the agent, builds the digital twin, generates scenarios and evals, runs them, and proposes fixes. Runs can also go headless in CI, and production traces can be pulled in through LangSmith, Langfuse, Arize or Braintrust so real failures become replayable scenarios.
The vendor also sells post-training, a separate product that turns simulation data into task-specific small models. The plans on this page cover the simulation product only; post-training has its own usage pricing on the same pricing page.
Source: Pricing | Plurai Simulation & Post-training.
Key features
Digital twin of the agent's world
Plurai reads the agent's code and wires a mocked copy of the systems it calls, seeded with realistic records and state that change as the agent acts.
Scenarios that cover the flow graph
Plurai models the agent's flow graph and generates scenarios across it, including hard personas that push back and probe policies.
Failures traced to the exact code
Each failure is root-caused with evidence and the exact code location, and each proposed fix is verified against the full regression set.
Runs in CI or headless
Simulations run in an isolated sandbox on the developer's machine or headless in CI, and the world resets to the same state every run.
Source: Pricing | Plurai Simulation & Post-training.
Pricing
3 plans
Overage is off by default: an admin sets an overage limit to keep running past the included credits, and extra Sim credits are billed after the period ends
Strengths and limits
Strengths
- Starts free: the Free plan gives a one-time allowance of 12,300 Sim credits, 2 agent setups, 7 scenarios, 3 runs and 10 sandbox hours.
- One command in Claude Code, Codex or GitHub Copilot maps the agent, builds the digital twin, generates scenarios and evals, runs them, and proposes fixes.
- Every failure is traced to the exact code location with evidence, and each fix is verified against the full regression set before it counts as fixed.
- Runs stay isolated: every write to real systems is mocked by default, so the agent can refund, cancel or update records without touching real customers or data.
Limits
- Needs a coding agent: Claude Code, GitHub Copilot CLI or Codex CLI, with Cursor listed as coming soon, so there is no standalone web app to run a simulation from.
- The free allowance is one-time only: once the 12,300 credits, 3 runs and 10 sandbox hours are used, simulations pause until a paid plan is chosen.
- Usage is metered in Sim credits, one cent of model-provider usage each, so long conversations cost more than short ones.
- Public rating is thin: 5.0 out of 5 stars from 2 reviews on Product Hunt, with no listing on G2 or Capterra.
Based on Pricing | Plurai Simulation & Post-training.
Who it is for
A good fit if you
- test an AI agent before real users meet it
- find policy violations, hallucinations and broken tool calls in a sandbox
- replay production failures from the exact turn they broke
- run agent checks in CI on every pull request
Look elsewhere if you
- want to test an agent without a coding agent installed
- want a fixed monthly price with no usage metering
- want a large body of public user reviews before choosing
Based on Pricing | Plurai Simulation & Post-training.
Common questions
Which coding agents does Plurai Simulation work with?
Claude Code, GitHub Copilot CLI and Codex CLI, with Cursor listed as coming soon. A browser sign-in is needed once, and the plugin handles the rest.
How does Plurai Simulation build the simulated world?
It reads the agent's code, maps every tool and dependency, and builds a digital twin of the systems the agent calls, seeded with realistic records and state, plus synthetic users and multi-turn scenarios.
Can Plurai Simulation test an agent before its first real user?
Yes. It builds the digital twin, scenarios and evals from the agent's code alone, so no real traffic is needed. Production traces can be connected later to ground the suite in real behavior.
How are failures fixed?
Each gap comes with a diagnosis pointing to the code that caused it and a proposed fix with evidence. Fixes are applied one at a time with approval, then re-run against the scenarios that exposed each gap, and a gap kept as a regression test is caught if it comes back.
Does Plurai Simulation touch production data?
No. Every write to real systems is mocked by default, so the agent can act freely without affecting real customers or data. A live write happens only if explicitly requested.
Prices from each tool's listed source.
Ready to look at Plurai Simulation?
Free plan, from $60/mo
Sources: Pricing | Plurai Simulation & Post-training (Simulation tab prices: Free $0, Base $60, Business $210 a month. Post-training tab is a separate product.), AI Agent Simulation & Testing Platform | Plurai, FAQ | Plurai, Plurai on Product Hunt (5.0 out of 5 stars from 2 reviews.), Plurai | The Recursive Self-Improvement Layer for AI Agents. The star rating is the public rating on the store the tool is listed on. How we rate