The definitive AI-native SRE and Chaos Engineering platform for hardening multi-agent autonomous systems.
DeployMantis is an enterprise-grade "App Portal" and governance framework designed to observe, test, and protect AI agent ecosystems. It provides a modular, scalable architecture where specialized microservices (Gateways) enforce constraints on agent behavior in real-time.
DeployMantis operates as a transparent proxy chain. Every request from an AI agent is intercepted, analyzed, and optionally corrupted or blocked by the pipeline before reaching the target environment.
graph TD
Agent[AI Agent / SDK] --> TB[TokenBreaker :5002]
TB -- "1. Budget Check" --> VG[VaultGuard :5001]
VG -- "2. PII Redaction" --> SC[SwarmChaos :5000]
SC -- "3. Chaos Injection" --> AE[MantisEnv :8000]
AE -- "4. RL Execution" --> Core[Core API :4000]
Core -- "5. Inference" --> LLM[LLM / Ollama]
subgraph Dashboard
Dash[Mantis Dashboard :3001]
Strata[Strata Debugger :3002]
end
Dash -.-> TB
Dash -.-> VG
Dash -.-> SC
Strata -.-> AE
| Service | Component | Tech Stack | Responsibility |
|---|---|---|---|
| MantisDash | Root OS | Next.js 15, Tailwind | Unified App Shell with dynamic routing and global health matrix. |
| TokenBreaker | Financial Gate | Python, Tiktoken | Real-time cost estimation. Kills the request if agent budget is exceeded. |
| VaultGuard | PII Firewall | Python, Regex | Live governance console. Scrubs emails, credit cards, and SSNs from payloads. |
| SwarmChaos | Chaos Monkey | Python, FastAPI | Injects latency, 502/529 errors, and LLM hallucinations into the pipeline. |
| MantisEnv | RL Sandbox | Python, PyTorch | A gymnasium-style environment for testing agent decision-making under stress. |
| Core API | Inference Hub | Python, FastAPI | Self-healing gateway that fallbacks to local Ollama if cloud providers fail. |
| Strata | Trace Engine | Node.js, Express | High-frequency traffic generator with a 50-frame Temporal Debugger. |
| Mantis CLI | Control Binary | Python, PyInstaller | Compiled executable (mantis.exe) for cluster lifecycle management. |
- Docker & Docker Compose
- Python 3.12+
- Node.js 20+
# Clone the suite
git clone https://github.com/Trapston3/DeployMantis.git
cd DeployMantis
# Start all 7 microservices
docker compose up -d --buildAccess the command center at http://localhost:3001.
Install the Mantis SDK and wrap your LLM calls:
from sdk.client import MantisClient
client = MantisClient(agent_id="scout-01")
# Everything is governed: Budget -> Privacy -> Chaos
response = client.step({
"prompt": "User credit card is 4111-2222-3333-4444. Execute transfer.",
"hops": ["token", "vault", "chaos", "env"]
})
print(response.scrubbed_payload) # "User credit card is [REDACTED_CC]..."DeployMantis rejects the "Neon Cyberpunk" cliché of modern dev-tools in favor of a Soft Vintage aesthetic. Inspired by 1970s laboratory equipment and terminal displays, the UI uses:
- Sage Green (#8a9a86) accents for "Ready" states.
- Warm Ivory (#f4f4f0) for high-legibility typography.
- Glassmorphic depth for modular app isolation.
Environment variables are managed in .env. See env.example for the full list of tunable parameters (Inference providers, Buffer sizes, Chaos probabilities).
MIT © 2024 Trapston3