Vertex Palace

The description states that Vertex Palace is a local memory and context router for Codex and other coding agents. It maps repositories into structured "palaces" with floors, rooms, drawers, symbols, dependencies, tests, routes, project memories, and a Pitfall Board.

It supports:

• Scanning TypeScript, JavaScript, Python, Markdown, JSON, and fallback formats.
• Indexing symbols, imports, dependencies, tests, directories, hashes, and architectural relationships.
• Classifying tasks and selecting adaptive operating modes (bypass, route-lite, full-palace, guarded-memory-palace).
• Separating context into Primary, Support, Deferred, and Excluded evidence.
• Preserving project decisions and previous mistakes as scoped memory.
• Adding execution boundaries, required evidence, stop conditions, and warnings.
• Evaluating files changed after a task and reporting route coverage or overconfidence.
• Running through CLI, MCP server, or Codex plugin.

It is described as not simply a vector-search RAG system, but one that combines deterministic repository structure, task intent, dependency relationships, scoped memory, explicit exclusions, and post-task evaluation.

The product is built in TypeScript and Node.js, with an open-source monorepo architecture. It includes scanners, parsers, structural indexes, adaptive route planners, context packers, memory systems, evaluators, CLI, MCP server, Codex plugin, fixture repositories, and automated tests.

The author states that the inspiration came from transforming two-dimensional information into a three-dimensional mental structure, leading to the human memory-palace technique. The goal was not to give an AI agent more context, but the right context at the right time.

Initially, the project aimed to reduce token usage and execution time by providing bounded context. However, research revealed that smaller routes do not automatically lead to efficiency gains — in fact, some experiments showed increased token use and tool calls compared to baseline methods.

The positioning has evolved from a performance-focused tool to one emphasizing:

• Structured repository routing
• Historical decision preservation
• Prevention of repeated mistakes
• Client and tenant guardrails
• Auditable task context

The author now emphasizes that confidence is not evidence, and that route quality must be tied to task-intent coverage, required evidence, exclusions, and actual file changes.

Not evidenced. The description does not identify specific customer segments or personas. It describes the tool's utility for AI coding agents like Codex but does not name target users or organizations.

The author notes that it supports Windows, macOS, and Linux with Node.js 20 or newer — suggesting a developer audience — but no explicit ICP is stated.

Not evidenced. There is no mention of pricing models, monetization strategies, or business structures beyond the open-source nature of the project.

The author states that Vertex Palace is open source and available via npm, with no indication of paid services or commercial offerings.

The product is described as:

• A TypeScript and Node.js monorepo
• Built using GitHub Actions, GPT-5.6 (as a co-engineer), model-context protocol (MCP), OpenAI Codex, pnpm, Vitest
• Supporting CLI, MCP server, and Codex plugin integrations
• Including scanners, parsers, structural indexes, adaptive route planners, memory systems, evaluators, fixture repositories, and automated tests

It has:

• A stable public npm package at v0.2.3
• A source release candidate at v0.3.0 (not yet published)
• CI coverage on Windows, macOS, and Ubuntu with Node.js 20 and 24
• 90 passing workspace tests at the latest product gate
• Exact routing gates on Zod TypeScript monorepo and Requests Python repository
• A reproducible benchmark repository containing protocols, raw artifacts, analysis, limitations, and stopped experiments

Not evidenced. The description does not contain any data about:

• Revenue
• Customers or users
• Adoption rates
• Market traction
• Product usage metrics

The author reports on controlled studies and research outcomes, including 124 valid and correct arm executions across v1 and v2.2 studies, but these are internal benchmarks, not external adoption.

Not evidenced. No mention of competitors or competitive landscape is provided in the description.

• No real-world usage or traction: The project appears to be an engineering experiment with no evidence of customer adoption.
• Unclear commercial viability: The tool is open-source, and there's no indication of monetization plans or business model.
• Research outcomes are mixed: While the system can measure its own weaknesses, it has not yet proven consistent performance improvements over baseline methods.
• Dependency on author’s own testing: All validation comes from internal studies; no independent verification or external feedback is reported.
• Lack of clear value proposition for end users: The tool is designed for AI agents, not directly for developers, so its utility beyond research remains unclear.

1. What specific use cases have you identified for Vertex Palace outside of your own testing?
2. Have you conducted any user feedback sessions or interviews with developers using AI coding tools?
3. How do you plan to transition from a research prototype to a scalable product or service?
4. Are there any plans to monetize the tool, and if so, what is the business model?
5. What are the key technical challenges in scaling this system for enterprise-level repositories?
6. How do you intend to validate performance improvements beyond controlled studies?