Show HN: Pacific Slate: a self-hosted, model-agnostic multi-agent AI assistant
Pacific Slate is a self-built, open-source personal AI system that autonomously updates from curated sources, filters noise, and maintains a privately owned encrypted database on a rented server It employs a multi-agent tree architecture on Google's Agent Development Kit (ADK) with one routing operator and seven specialized agents (coder, researcher, analyst, productivity, reviewer, evaluator, and a research sub-agent), each running on a model chosen for cost and capability fit The system priori
Analysis
TL;DR
- Pacific Slate is a self-built, open-source personal AI system that autonomously updates from curated sources, filters noise, and maintains a privately owned encrypted database on a rented server
- It employs a multi-agent tree architecture on Google's Agent Development Kit (ADK) with one routing operator and seven specialized agents (coder, researcher, analyst, productivity, reviewer, evaluator, and a research sub-agent), each running on a model chosen for cost and capability fit
- The system prioritizes durability and utility over novelty, using MCP/plugins/hooks for integration rather than custom clients, with a service-agnostic design that treats the model as a swappable commodity
- All data stays on the user's server, requests route to zero-data-retention endpoints configured not to train on or publish data, and the architecture was built AI-natively with coding agents writing most of the code under human specification and review
- Production operation since early 2026 revealed design gaps rather than bugs, leading to pragmatic additions like a fallback layer and per-answer model labels for behavioral regression tracking
Why It Matters
Pacific Slate represents a growing shift toward personal, self-hosted AI systems that prioritize data sovereignty, cost transparency, and workflow integration over proprietary cloud services. For AI practitioners, it demonstrates a practical multi-agent architecture that balances specialization with economic efficiency by mapping roles to appropriately sized models. The system also serves as a case study in AI-native development, showing how non-programmers can architect and operate complex agent systems through iterative, operation-driven refinement.
Technical Details
- Architecture: Multi-agent tree built on Google's Agent Development Kit (ADK), featuring one root operator agent that routes requests to seven specialist agents (coder, researcher, analyst, productivity, reviewer, evaluator, and a research sub-agent scoped to the coder), with each role mapped to a model selected for cost-performance fit
- Data & Privacy: All accumulated data resides on a privately rented server with encrypted backups; outgoing requests are restricted to zero-data-retention endpoints at providers configured (as of August 2026) not to train on or publish user data; local models can handle sensitive work that never leaves the server
- Integration & Interface: Reaches users through MCP, plugins, and hooks embedded in existing tools rather than custom clients; answers are delivered as movable cards labeled with model used, cost, and latency; a live dashboard (Monitor) displays seismic data, headlines, and sample personal data
- Development Approach: Built AI-natively with the architect specifying and reviewing while coding agents wrote most of the code; the design emerged iteratively from production operation rather than upfront planning, with pragmatic fixes like a custom fallback layer and model provenance labels added in response to real-world failures
- Open Source & Cost Model: Nearly all components are open source and self-hosted; the only paid dependencies are the rented models and one swappable hosted memory service that is explicitly not the system of record
Industry Insight
- The trend toward personal, self-hosted AI systems will accelerate as professionals seek data sovereignty and cost control, making architecture patterns like Pacific Slate's model-agnostic, service-swappable design increasingly relevant as reference implementations
- The AI-native development approach demonstrated here—where non-programmers architect systems and agents write code—signals a lowering of barriers to entry for complex multi-agent deployments, though it underscores the continued importance of human oversight in design and review
- The emphasis on production-driven iteration over upfront planning reveals a key lesson for practitioners: multi-agent systems often expose design gaps in operation that cannot be predicted in simulation, making robust logging, model provenance tracking, and graceful degradation essential features from the start
Disclaimer: The above content is generated by AI and is for reference only.