Your AI Stack, finally
under control
secure
compliant
budget-safe
telemetry-free
A self-hosted gateway that masks secrets, gates code quality, and kills runaway token spend — before any of it becomes your problem.
For infra-paranoid engineering leaders who still want to ship agents fast.
The Threat Feed
Your team is using AI. Your infrastructure has no visibility.
Logs are leaking, token budgets are ignored, and buggy AI code is piling up. This is your stack right now:
Services
Nine services. One pipe. Nothing leaks.
DeployMantis intercepts, sanitizes, and controls all outbound AI traffic through specialized local nodes.
Vault-Guard
Real-time regex scanner that redacts API keys, credentials, and PII before prompts leave your network. Fully customizable via YAML rules.
MantisVerify
Gating service that scores every AI-generated code change before humans review it. Analyzes AST, checks structure, and marks PRs as PASS or FAIL.
- eval(user_input) # risky
+ safe_execute(user_input)
def safe_execute(code):
# AST validation added
- No dynamic evals found
- AST safety nodes verified
- Execution boundaries matching
Token-Breaker
Financial circuit breaker that enforces strict budget caps per developer and team. Runaway loops are terminated instantly.
Fallback-Mesh
Ensures service availability. If primary cloud models (e.g. OpenAI) fail or rate-limit, Fallback-Mesh automatically routes payloads to pre-configured local Ollama models.
Strata Tracing
Every token, latency metric, and request audit trail is logged and exported locally. Odysseus dashboard visualizes service logs and team usage statistics.
MantisSnap
Branch-aware context restoration. Save your working memory state to local SQLite cache when switching branches. Instantly reload TODO logs.
Switched to branch 'feat/scanner'
$ curl http://localhost:5004/context
{ "last_active_file": "vault-guard/main.py", "todos_pending": 3 }
Core Infrastructure Services
core-api
FastAPI-powered central routing layer. Handles asynchronous validation, telemetry extraction, and coordinates microservice pipelines with <5ms overhead.
Odysseus Dashboard
Central dashboard for microservice health, budget forecasting, and configuration. Live-tweak chaos injection rates and view audit logs in real time.
SwarmChaos
Chaos testing orchestration. Inject latency, simulated timeouts, API errors, and synthetic model hallucinations to test the self-healing ability of your agent clusters.
Setup
docker compose up. That's it. Everything else is already configured.
DeployMantis runs entirely on your local infrastructure. Zero data leaves your system.
Clone & configure
Clone the git repository and copy variables into your local .env file.
Launch gateway
Run compose up to launch all nine gateway services in isolated Docker containers.
Redirect API base
Change your LLM API endpoint to localhost:4000. Intercept begins immediately.
Verify logs
Open local dashboard at localhost:3000 to monitor alerts and budget usage.
CLI Setup
# 1. Clone the repo git clone https://github.com/trapston3/deploymantis cd deploymantis # 2. Configure environment cp .env.example .env # 3. Boot local stack docker compose up
Pricing
Start free. Pay when you ship.
No credit card needed to get started. Self-hosted means your infra, your data, your control.
Solo
For individual developers. Self-hosted, no time limit.
- Full gateway + all 9 services
- Vault-Guard secret masking
- Token-Breaker budget tracking
- Runs happily on a single dev laptop
- MantisVerify quality gate
- Team seats & audit logs
- Priority support
Team
For teams that need CI quality gates, per-seat budgets, and a compliance export before your next audit.
- Everything in Solo
- MantisVerify AI quality gate
- Trust Score per response
- Team-scoped budgets & alerts
- Audit log & compliance export
- MantisHeal self-healing CLI
- Email support (48h SLA)
First 50 seats locked at $9/mo
Enterprise
For orgs that need SLAs, custom policy engines, and white-glove onboarding.
- Everything in Team
- Custom secret scanning policies
- SSO & SCIM provisioning
- On-prem deployment support
- Dedicated Slack channel
- Custom SLA & legal
- Quarterly security reviews
Audit Logs
For teams where 'move fast' stopped being funny after the incident.
"We routed our agent cluster through Vault-Guard. It caught three raw AWS session tokens in developer contexts before they reached external endpoints."
"Our CI gates changed when we added MantisVerify. Instead of reviewing 500-line generated PRs manually, we block any diff where the AST quality score drops below 0.8."
"A developer left a recursive loop running on an experimental Claude agent overnight. Token-Breaker tripped the circuit breaker after $20 of spend, saving us from a four-figure bill."
FAQ
Common questions
No. DeployMantis runs entirely on your servers. The only outbound calls are the ones you configure — to LLM providers you choose. No telemetry, no phone-home.
Any LLM with an OpenAI-compatible API — including OpenAI, Anthropic, Mistral, Groq, Ollama, and any local model via LM Studio or text-generation-webui.
Docker + 4GB RAM for all 9 services. Tested on Ubuntu 22.04, macOS 14+, and Windows 11 WSL2. A single developer laptop works fine for solo use.
Yes — the gateway, secret scanner, and observability layer are MIT-licensed on GitHub. Paid tiers unlock MantisVerify, Team features, and priority support.
Team pricing is per-seat. For teams over 50 seats, contact us for volume discounts. Enterprise contracts include fixed annual pricing with no per-seat caps.
Yes. DeployMantis uses a standard OpenAI-compatible proxy interface. Point your existing agent or IDE plugin at localhost:4000 and it works — no SDK changes required.
50 teams. No cloud. No telemetry. No nonsense.
Secure your AI integration locally. Claim permanent early access pricing.
You're on the list!
We've received your request. We will reach out to you at within 48 hours.
Self-hosted gateway · Open-source core