ArchAI - Data Broker

The description states that ArchAI is:

• A lightweight, database-free MCP service.
• Built in Node.js, using MCP over standard input and output.
• Designed to remove sensitive information (API keys, tokens, emails) from outgoing context.
• To organize context into traceable layers: Identity, Instructions, Task, Memory, Evidence, Recent activity.
• To enforce configurable token budgets using a formula: $\hat{T}(x)=\left\lceil\frac{|x|}{4}\right\rceil$.
• To route requests between Ollama, vLLM, and OpenAI-compatible providers.
• To produce metadata-only audit events, without storing user conversations.
• To expose six tools: archai_redact, archai_context, archai_route, archai_status, archai_ask, archai_audit.
• Compatible with any MCP-capable application without requiring internal storage adoption.

Inferred: It is a control layer, not an AI model or inference engine itself. It does not store data, but it tracks what was sent and why.

The description states:

• ArchAI positions itself as a privacy-first MCP layer.
• It aims to create transparency in how AI clients interact with models.
• It emphasizes user control over privacy, context, routing, and audit data.
• It claims to not own user data, instead offering redaction, traceability, and routing.

Inferred: The positioning evolved from a hackathon prototype into a tool for developers seeking control and auditability in AI workflows. The claim is that it allows switching between local and cloud providers without changing the application contract.

The description states:

• ArchAI is built for MCP-capable applications, which implies developers or integrators working with AI clients.
• It targets users who want to use Ollama, vLLM, and OpenAI while maintaining control over data flow.
• It is designed for those who want to avoid surrendering ownership of their data or database.

Inferred: The ICP (Ideal Customer Profile) likely includes:

• Developers building AI applications using MCP.
• Teams seeking privacy compliance, auditability, and flexibility in model runtime choices.
• Users who are not interested in proprietary solutions but want interoperable control layers.

Not evidenced: No specific customer segments, personas, or use cases beyond the hackathon submission.

The description states:

• ArchAI is a self-contained service, built with no database.
• It is described as a lightweight tool for developers.
• It does not mention any pricing model, monetization strategy, or commercial offering.

Inferred: The business model is likely not yet defined. It may be a developer tool or open-source project that could evolve into a SaaS or consulting offering later.

Not evidenced: No revenue, pricing, or monetization details.

The description states:

• ArchAI is implemented in Node.js, using MCP over standard input/output.
• It requires no database, background telemetry, or inbound network port.
• Context is divided into ordered layers: Identity, Instructions, Task, Memory, Evidence, Recent activity.
• It uses a formula to estimate token cost: $\hat{T}(x)=\left\lceil\frac{|x|}{4}\right\rceil$.
• It selects context layers based on a token budget constraint.
• It exposes six MCP tools for redaction, routing, status, and audit.
• It supports metadata-only audit events, without storing conversations.

Inferred: The technical architecture is lightweight, interoperable, and privacy-focused. It avoids traditional data storage or complex infrastructure dependencies.

Not evidenced: No information on scalability, performance, or production readiness.

The description states:

• ArchAI was built for the OpenAI 2026 hackathon.
• It is described as a self-contained service, not yet a product with customers or revenue.
• The authors emphasize that it was built quickly and tested in a constrained environment.

Inferred: This is a prototype or early-stage tool, likely not yet adopted by users or integrated into production workflows.

Not evidenced: No customer data, usage metrics, or adoption signals beyond the hackathon submission.

The description states:

• ArchAI aims to solve issues around privacy, provenance, token management, and routing.
• It is positioned as a control layer for AI applications using MCP.
• It supports local runtimes (Ollama, vLLM) alongside cloud providers (OpenAI).

Inferred: The competitive space includes:

• Other MCP-compatible tools or middleware.
• Privacy-focused AI control layers or data governance tools.
• Tools that enable switching between local and cloud inference engines.

Not evidenced: No direct competitors named, no market size, or competitive positioning data.

The description states:

• ArchAI is a hackathon submission, not yet a mature product.
• It does not store user conversations, but it does report what was removed and hashes payloads.
• It uses pattern-based redaction, which the authors note is not a substitute for full security review.

Inferred:

• Risk: Limited real-world testing or production use.
• Risk: Pattern-based redaction may not be sufficient for high-security environments.
• Risk: No database or persistence layer may limit auditability or long-term tracking in complex workflows.
• Open question: The project is not yet commercialized, and no evidence of traction or monetization.

1. What is the current status of ArchAI beyond the hackathon? Is it being used by any developers or applications?
2. How does ArchAI handle edge cases in redaction, especially with complex or obfuscated data?
3. Are there plans to support additional providers or model formats beyond Ollama, vLLM, and OpenAI?
4. What is the intended monetization strategy for ArchAI? Is it open-source or SaaS?
5. How does ArchAI integrate with existing MCP clients in production environments?
6. Has the team considered performance implications of token budgeting at scale?