IgnisPrompt: Building an Auditable, Zero Cloud Dependency Local AI Routing Infrastructure

IgnisPrompt is a local-first AI routing infrastructure project focused on providing reviewable and auditable AI workflow support. By default, it has zero cloud calls and zero telemetry sending, offering enterprises and developers a fully controllable AI inference routing solution. The project is written in Rust, with the current version being v0.1.3-local-preview, addressing data privacy, compliance auditing, and vendor lock-in issues faced by enterprises when using LLMs.

With the popularization of LLMs in enterprise scenarios, enterprises face the contradiction of wanting to leverage AI capabilities while worrying about data privacy, compliance auditing, and vendor lock-in. Traditional AI solutions rely on cloud services, requiring sensitive data to leave the local environment, and enterprises lose full control over the inference process. IgnisPrompt was born to address this, with the core concept of local-first: by default, zero cloud calls, zero telemetry, and zero global aggregation, allowing enterprises to run AI workflows in a controlled environment and meet compliance audit requirements.

IgnisPrompt is a local AI routing daemon written in Rust, providing an HTTP control plane to manage inference requests. Its core features include:

1. Zero Cloud by Default: No data sent to cloud services without explicit configuration
2. Fully Auditable: Locally records routing decisions, audit events, and evidence
3. Adversarial Content Security: Detects and marks potential adversarial documents
4. Modular Design: Supports runner providers like local GGUF models Key components include the ignispromptd daemon, ignispromptctl command-line tool, Aethra module (local contract enhancement), and MCP stub (observability). The routing mechanism is transparent: analyze request → decide routing → generate explanation → record audit → generate evidence package.

The tech stack uses Rust, with advantages including memory safety, high performance, concurrency safety, and portability. API endpoints comply with OpenAI-compatible formats, such as /health (health check), /v1/models (model list), /v1/audit/events (audit event query), etc. Support for local GGUF models: requires configuring the runner path, providing .gguf weight files, enabling timeout and pre-check enhancements to achieve fully offline inference.

• Enterprise Compliance Scenarios: Industries like finance and healthcare can ensure data sovereignty (sensitive data retained locally), audit tracking (complete logs and evidence chains), and vendor independence (not tied to cloud service providers)
• Development and Testing Scenarios: Stub runner and smoke test support rapid prototyping, CI/CD friendliness, and single-command development environment startup
• Research and Education Scenarios: Transparent decision explanations, local evidence packages, and adversarial testing mechanisms help with AI interpretability and audit research

Known Limitations: Not yet reaching production-level model quality, no enterprise compliance certification, streaming and MCP support are experimental, no packaging and distribution mechanism, and design partner readiness needs improvement. Version Evolution: v0.1.0 (technical MVP) → v0.1.1 (local preview) → v0.1.2 (patch) → v0.1.3 (added security review, contract enhancement, etc.). Community Contributions: Contributions are welcome, with detailed guidelines (branch naming, commit signing, PR templates, etc.) provided.

IgnisPrompt represents an AI technology trend: leveraging AI capabilities while maintaining control over data and processes. Its local-first, auditable architecture provides an attractive option for organizations under compliance constraints. Although in the early stages, the clear architecture and honest evaluation demonstrate the team's maturity. It is recommended that technical decision-makers focusing on AI governance and data privacy continue to pay attention to this project. Its success depends on community feedback and accumulated deployment experience, and it may become an important part of enterprise AI architecture.