Archal provides a pre‑deployment staging platform that runs AI agents against stateful digital twins of production services such as GitHub, Slack, and Linear. Users author test scenarios in markdown, and the platform provisions the twins, executes agents via MCP, captures full execution traces, and generates automated probabilistic satisfaction scores. This enables AI product teams and DevOps engineers to validate agent behavior safely before release.
Funding
Funding not disclosed
Founders
Product
Problem
AI agents that interact with production tools (e.g., GitHub, Slack, Linear) often lack a safe testing layer, causing unexpected API calls, hallucinations, or off‑script actions to surface only after deployment. Without realistic sandbox environments, teams must rely on ad‑hoc debugging in production, leading to service disruptions and costly rollbacks.
Solution
Archal provides a pre‑deployment staging platform that lets developers run AI agents against stateful digital twins—exact behavioral clones of the production services they will consume. Users author test scenarios in markdown, specifying initial state, expected behavior, and success criteria. The platform provisions the twins, executes the agent via MCP‑compatible interfaces, and captures a full execution trace, including every tool call and decision point. An automated evaluator generates a probabilistic satisfaction score and detailed per‑criterion results, enabling rapid iteration and red‑team testing. All traces are stored securely and can be inspected through a web dashboard or CLI, allowing teams to validate agent logic before any real‑world impact.
Target Audience
The primary users are AI‑driven product teams, DevOps engineers, and platform developers who build autonomous agents that call external services, as well as enterprises seeking to certify agent reliability before production rollout.
Features
- **Digital Twin Engine**: Stateful, high‑fidelity clones of external services (GitHub, Slack, etc.) that behave like production APIs without side effects.
- **Markdown Scenario DSL**: Structured test definition language for setup, actions, and success metrics, supporting deterministic and probabilistic criteria.
- **Trace Capture & Replay**: End‑to‑end logging of every API request, response, and agent decision, stored in a searchable, immutable audit log.
- **Automated Satisfaction Scoring**: Probabilistic evaluation model that quantifies how well the agent meets scenario expectations.
- **CLI & Dashboard Access**: Integrated @archal/cli for CI pipelines and a web UI for interactive review, both supporting API‑key authentication and role‑based access.
- **MCP Compatibility**: Seamless integration with existing model‑control‑protocol interfaces for plug‑and‑play agent execution.
- **Secure, Encrypted Storage**: HIPAA‑grade encryption at rest and in transit, with granular permission controls.
- **Scalable SaaS Backend**: Multi‑tenant architecture that provisions twins on demand, supporting both cloud‑hosted and Docker‑based local execution.