Concept
Perstack is a containerized harness for agentic apps. Three layers of design — philosophy, orchestration model, and harness architecture — shape how it works.
The name “Perstack” combines the Latin word “perītus” (meaning “expert”) with “stack”. Perstack = Expert Stack.
Design philosophy
Section titled “Design philosophy”Perstack is built on a single thesis: output quality is determined by the strength of verification signals in the loop, not by the capability of the model.
When an agent generates output, something checks whether that output is correct. That check produces a signal. Signals come in two kinds:
- Soft signals depend on LLM judgment — requirements alignment checks, semantic review, ad hoc manual testing. These signals are valuable for qualitative evaluation that only an LLM can provide, but cannot serve as the final gate — the verifier’s judgment can vary across runs.
- Hard signals are independent of LLM judgment — compiler errors, E2E test results, screenshot diffs. These signals are binary, deterministic, and unambiguous. Agents looping on hard signals detect breakage reliably and self-correct.
Every architectural decision in Perstack — context isolation, deterministic runtime, sandbox boundaries, full observability — exists to maximize hard signal opportunities. Context isolation ensures the verifier never sees the generator’s reasoning (context separation). Deterministic state ensures the same verification always produces the same result (determinism). The workspace provides a controlled space where actual artifacts can be built and tested against ground truth.
This is what “quality is a system property” means in practice. The harness provides the verification floor — hard signals that guarantee the output works. The Expert author provides the knowledge ceiling — domain constraints that ensure the output solves the right problem. Quality emerges from the combination, not from model capability or architecture choice.
For the full framework — the three conditions that make a signal hard, how Perstack’s architecture maps to each, and practical guidelines for designing verifiable Experts — see Hard Signals. For a real-world example, see
create-expert— Perstack’s own Expert for generating Expert definitions, which embeds a hard signal verifier in its delegation tree.
Expert Stack
Section titled “Expert Stack”Three principles guide how Perstack approaches agentic app development:
- Quality is a system property, not a model property — Building agentic apps people actually use doesn’t require an AI science degree — just a solid understanding of the problems you’re solving.
- Keep your app simple and reliable — The harness is inevitably complex — Perstack absorbs that complexity so your agentic app doesn’t have to.
- Do big things with small models — If a smaller model can do the job, there’s no reason to use a bigger one.
Micro-agents — a multi-agent orchestration design
Section titled “Micro-agents — a multi-agent orchestration design”Perstack introduces micro-agents — a multi-agent orchestration design built around purpose-specific agents, each with a single responsibility.
- Simple — A monolithic agent assembles its system prompt from hundreds of fragments. A multi-agent framework stacks abstraction layers and wires orchestration in code. A Perstack expert is one TOML section — instruction, delegates, done.
- Reliable — A plan agent that only plans, a build agent that only builds, a verify agent that only verifies — the pipeline structure itself prevents shortcuts and catches errors that a single generalist would miss.
- Reusable — Delegates are dependency management for agents — like npm packages or crates. Separate concerns through delegate chains, and compose purpose-built experts across different projects.
Expert Stack — harness architecture
Section titled “Expert Stack — harness architecture”Perstack ships a five-layer stack that gives micro-agents everything they need to run.
┌──────────────────────────────────────────────────────────────────-┐│ Interface ││ CLI · Event streaming · Programmatic API │├──────────────────────────────────────────────────────────────────-┤│ Runtime ││ Agentic loop · Event-sourcing · Checkpointing · Tool use │├──────────────────────────────────────────────────────────────────-┤│ Context ││ System prompts · Prompt caching · AgenticRAG · Extended thinking │├──────────────────────────────────────────────────────────────────-┤│ Definition ││ Multi-agent topology · MCP skills · Provider abstraction │├──────────────────────────────────────────────────────────────────-┤│ Infrastructure ││ Sandbox isolation · Workspace boundary · Secret management │└──────────────────────────────────────────────────────────────────-┘Interface
Section titled “Interface”The top layer exposes Perstack to users and applications.
Runtime
Section titled “Runtime”The execution engine that drives the agentic loop.
Context
Section titled “Context”What each expert sees and knows during execution.
Definition
Section titled “Definition”How you declare what experts do and how they collaborate.
Infrastructure
Section titled “Infrastructure”Isolation and security boundaries that make agents safe to operate.