Shiliu

What the company appears to be: Shiliu is a self-reported infrastructure layer for AI agents that manages long-term context. It positions itself as a model-agnostic memory service that governs what information should persist or be suppressed across multiple AI tools, without replaying full conversation history.

What changed: The project description indicates an evolution from building chat interfaces to solving the underlying infrastructure problem of shared, governed context for AI agents. This shift suggests a move toward platform-level thinking rather than product-level solutions.

Single most important open question: Is there evidence of real-world usage or integration beyond the demo and benchmarking? The description states that Shiliu is "a live product site" but provides no details on adoption, customers, or revenue — only claims about technical capabilities and benchmarks.

The description states that Shiliu is a model-independent long-term context layer for AI applications. It sits between an application and its model call, managing what information should be shared with agents without replaying full conversation history.

It exposes three main API endpoints:

• POST /v1/turns/ingest processes completed conversation turns.
• POST /v1/memory/recall retrieves governed memory and diagnostics.
• POST /v1/chat/context creates a neutral context capsule for an agent.

Shiliu is built around:

• A MemoryService core
• Effective Memory View policy ordering
• Multiple recall lanes
• Change Trace
• Evidence anchors
• Identity and scope resolution
• Persistence adapters (SQLite for local, PostgreSQL for production)

The system supports tenant, user, application, agent, persona, and visibility scope separation.

Inference: The product appears to be a memory management platform designed to solve state alignment across multiple AI agents. It is not a chatbot or standalone tool but an infrastructure component.

The description states that Shiliu was built to solve the infrastructure problem underneath AI chat interfaces, rather than building another chat UI.

It claims:

• “We did not want to build another chat interface.”
• “Shiliu is a model-independent long-term context layer for AI applications.”
• “Shiliu returns context rather than the final chatbot reply, so the host agent keeps control of reasoning, personality, and expression.”

This positioning suggests a shift from product-focused to platform-focused thinking — aiming to become foundational infrastructure for AI agents.

Inference: The company evolved from a tool-building mindset to an infrastructure-building one. This is a common transition in early-stage AI startups, but the description does not indicate whether this evolution has led to real traction or customer feedback yet.

The description does not clearly define target customers or ideal customer profiles (ICP). It mentions that Shiliu supports:

• Agents
• Applications
• Models
• Workflows

It also notes support for:

• Tenant, user, application, agent, persona, and visibility scope separation.

However, there is no mention of specific industries, roles, or use cases beyond AI agents using it as a memory layer.

Inference: The ICP likely includes developers building AI agents or platforms that need to manage shared context across multiple tools. But the description does not specify who those users are or how they would integrate with Shiliu.

There is no evidence in the description of a business model or pricing strategy.

The description states:

• “Shiliu itself will continue to develop as the independent context infrastructure underneath many agents.”
• “Next we will package the API and SDK for faster partner integration.”

It also mentions:

• “Customer-private data is not used for model training by default.”
• “Writes use trace IDs and idempotency controls, and clients receive explicit memory outcomes instead of treating an HTTP success response as proof that durable memory was created.”

These are technical features, not business or pricing details.

Inference: The business model remains unspecified. It may be SaaS-based, platform-based, or integrated into larger AI products — but no indication is given.

The description provides several technical signals:

• Built with Node.js, JavaScript, PostgreSQL, SQLite, REST APIs
• Uses Codex and GPT-5.6 for development
• Supports model-neutral API boundaries
• Implements explicit memory outcomes, deterministic policy ordering, evidence-aware diagnostics
• Includes Change Trace, Evidence Anchors, and scope resolution
• Has a public demo and regression testing (20 rounds, 100-turn corpus, 120-turn simulation, 500-question benchmark)
• Uses trace IDs and idempotency keys for persistence control

It also states:

• “A live product site and a model-independent API for ingestion, recall, context composition, governance, and diagnostics.”
• “The proprietary memory engine and production source code remain private; only a sanitized demo and integration surface are shared.”

Inference: The technical architecture is well-defined and shows strong engineering effort. However, the lack of real-world deployment or customer feedback limits its commercial relevance.

There is no evidence of traction or maturity beyond:

• A live product site
• Public demo and regression tests
• Benchmark results (LongMemEval-CN)
• A single-member team

The description mentions:

• “A live product site”
• “Public LongMemEval-CN evidence: a 499/500 first run on the published 500-question Chinese long-memory evaluation under the documented judge setup”
• “Broader private validation using a 100-turn Codex A/B cross-account memory corpus with six sampled cross-account recall probes”

But there is no mention of:

• Customers
• Revenue
• Usage metrics
• Product adoption

Inference: The project appears to be in an early stage, likely pre-revenue. It has technical validation but lacks commercial traction.

The description does not provide any information about competitors or competitive positioning.

It does not mention:

• Other memory management systems
• AI agent platforms
• LLM context window solutions
• Vector database providers

Inference: No competitive landscape is described. This makes it difficult to assess how Shiliu differentiates itself in the market.

Key risks and red flags based on the description:

1. No revenue or customer data: The project appears to be pre-revenue with no evidence of real-world adoption.
2. Single-founder team: Only one member is listed, which may limit execution capacity.
3. Unverified claims: All statements are self-reported and unverified — including performance benchmarks.
4. Lack of commercial traction: Despite technical sophistication, there’s no indication of real usage or integration beyond demos.
5. No pricing or monetization strategy: The business model is unclear.

Inference: The project is technically impressive but lacks commercial validation. It may be a promising idea in concept, but the absence of real-world use cases raises questions about viability.

1. What specific AI agents or platforms are currently using Shiliu?
2. How many active users or integrations exist beyond the demo and benchmarks?
3. What is the current monetization strategy, if any?
4. Are there any customers who have signed contracts or committed to paying for the service?
5. What are the key challenges in scaling this infrastructure for enterprise use?
6. How does Shiliu handle data privacy and compliance (e.g., GDPR, CCPA)?
7. Is there a roadmap for expanding beyond the current API and SDK offerings?
8. What is the plan for handling multi-tenant environments and governance at scale?