Agent Haderach

What the company appears to be: Agent Haderach is a self-reported project that describes itself as an infrastructure layer for coding agents. Its stated purpose is to provide persistent, searchable memory for AI coding agents working on repositories. It aims to store granular experience from agent sessions and make it available to future agents or developers.

What changed: The description indicates this is a hackathon submission (OpenAI 2026) with a complete implementation including backend services, frontend UI, MCP integration, and evaluation harnesses. The authors claim to have built a functional system that connects agents to a shared experience repository using structured summaries and retrieval mechanisms.

Single most important open question: Is there evidence of real-world usage or adoption beyond the hackathon prototype? The description states no revenue, customers or traction data are available — all claims are self-reported and unverified.

The description states that Agent Haderach is a system designed to give AI coding agents persistent memory. It connects to repositories through MCP (Model Context Protocol) and stores compact, evidence-backed records such as:

• workflows for building, testing, deploying, and debugging;
• lessons about architecture and code behavior;
• pitfalls describing failed approaches;
• summaries from completed or unfinished investigations;
• incidents affecting repository services;
• questions and verified answers.

It also includes a web application that allows developers to create workspaces, manage access, browse structured experience, inspect agent activity, and visualize knowledge movement through the repository.

The system is built as a TypeScript monorepo with:

• Next.js web application
• Hono service exposing REST endpoints and MCP endpoint
• PostgreSQL persistence with Drizzle
• Shared schemas and types
• Deterministic retrieval combining multiple signals

Not evidenced: The actual product functionality beyond the self-reported implementation details. No evidence of real usage or performance metrics.

The description states that Agent Haderach was inspired by the problem that "most of what they learn disappears when the session ends." It positions itself as a solution to the transition that application data once did — where information becomes large, dynamic, interconnected, and frequently queried, requiring infrastructure designed for structure, searchability, ranking, and reusability.

The authors claim it complements existing tools like Git history and repository documentation by preserving how previous agents learned to change code. They also state it creates collaboration between humans by making useful results of different approaches available across sessions and developer accounts.

They describe their approach as focusing on "durable, granular experience with explicit evidence, revision metadata, progressive retrieval, and feedback from actual reuse."

Not evidenced: No claims about market positioning beyond the hackathon context. No evidence of competitive differentiation or customer feedback that would validate these claims.

The description states that Agent Haderach targets developers working on repositories who use coding agents. It aims to help teams build on proven workflows, pitfalls, and knowledge from every previous task.

It also mentions that it connects developers as well as agents — preserving value generated by different engineering styles and agent setups without forcing everyone into the same workflow.

Not evidenced: No specific customer segments or personas identified. No evidence of target market size or early adopter profiles.

The description does not contain any information about pricing, monetization, or business model. It only describes the technical architecture and functionality of the system.

Not evidenced: No evidence of revenue streams, pricing models, or commercial arrangements.

The project is built as a TypeScript monorepo with several deliberately separated layers:

• Next.js web application
• Hono service exposing REST endpoints and MCP endpoint
• Optional stdio MCP transport for local agent integrations
• PostgreSQL persistence with Drizzle and SQL migrations
• Shared schemas and types used across frontend, API, and MCP boundary

It uses deterministic retrieval combining repository scope, metadata, lexical signals, evidence, confidence, freshness, and reuse feedback. It implements compact-first retrieval allowing agents to request full detail only for promising results.

The backend contains no LLM and makes no AI API calls — connected coding agents perform the judgment.

Not evidenced: No evidence of production deployment, scalability, or performance data beyond the hackathon evaluation harnesses.

The description includes accomplishments from a hackathon:

• Built a complete agent-to-agent experience loop rather than a static mockup
• Completed real-repository evaluations showing performance improvements (e.g., 45.6% fewer non-cached input tokens)
• Delivered a hosted multi-user product with repository workspaces, access control, personal MCP tokens, dashboard, and visualizations

However, the description explicitly states that no revenue, customer or traction data is available beyond what they state.

Not evidenced: No evidence of real-world usage, user base, or commercial traction. All claims are self-reported.

The description does not provide any information about competitors or competitive landscape. It only describes what the system does and how it was built.

Not evidenced: No evidence of existing solutions or competitive positioning in the market.

• Unverified claims: All statements are self-reported and unverified.
• No traction data: No evidence of real-world usage, customers, or revenue.
• Hackathon prototype: The system appears to be a hackathon submission with no indication of post-hackathon development or commercialization.
• Limited scope: The description focuses on internal tooling for developers using coding agents; unclear if it addresses broader enterprise needs.
• Dependency on agent adoption: Success depends heavily on widespread adoption of AI coding agents, which may not yet be mainstream.

1. What is the actual usage or adoption rate beyond the hackathon?
2. How does Agent Haderach plan to scale beyond a single repository?
3. Are there any existing partnerships or pilot programs with development teams?
4. What are the technical challenges in moving from a prototype to production-grade infrastructure?
5. How do you intend to monetize this product, and what is your go-to-market strategy?
6. What are the key assumptions about developer behavior that underpin your approach?
7. How does the system handle data privacy and security concerns for shared repositories?
8. What are the main technical limitations or bottlenecks currently in place?