Polycode v1: A Context-Window-Free Agentic Coding CLI

This is the release report for polycode v1, an open-source agentic coding command-line interface built by Polylogic AI. It describes what polycode is, where it came from, what it demonstrates, what it does not, and the v1.1.x release path from the initial v1.0.0 cut to v1.1.5. The companion academic paper, Compile: A Fourth Primitive for Canon-Keeping AI Agents (Salvo 2026d), formalizes the theoretical contribution and reports the fifteen-test validation harness. This report and the paper should be read together. Polycode operates without holding conversation state in the language model's context window. State lives in a SHA-256-chained, user-owned, append-only canon on the developer's local machine at ~/.polycode/canon/. The prompt passed to the language model on each turn is a structural query of that canon, bounded in size irrespective of how much history the canon has accumulated. This applies the type-separation architecture published in Engine, Rules, and Canon (Salvo 2026b) to the developer CLI surface for the first time, and in doing so, relaxes the coupling between agent coherence and context-window size that the current generation of agentic tools relies on. Every turn in polycode mints a witnessed commitment: a canon row containing the user's claim, the tool substrate it acted on, and a verdict produced by a deterministic non-LLM composer over the witness primitives. The composer's non-LLM property is the structural safeguard the paper proved necessary at n=30 in Salvo (2026c), where nine transformer reviewers from disjoint provider families agreed at Cohen's κ=1.000 on thirty NeurIPS 2024 papers, while the structurally non-transformer witness arm refused to commit on all thirty. Substrate, the paper concluded, cannot witness itself. Install: npx @polylogicai/polycode@latest on macOS, Linux, or Windows. Zero configuration, free hosted inference tier. Source: github.com/polylogicai/polycode (MIT) Package: npmjs.com/package/@polylogicai/polycode Disclosure. Polycode is inspired by Claude Code's public architecture at docs.claude.com. Polycode is not affiliated with Anthropic. No Claude Code code or system prompts are copied into polycode. Anthropic (compile tier) and Groq (generator tier) are consumed as arms-length paid API providers; neither has reviewed or endorsed this report. Chain: Derived from Salvo 2026b, continues Salvo 2026c, companion to the Compile paper (Salvo 2026d), supplemented by the polycode and polybrain-kernel repositories. PDF is Bitcoin-anchored via OpenTimestamps. License: CC BY 4.0 (this document). MIT (polycode source). Axiom: substrate cannot witness itself.

Authors

Institutions

Publication Details

Journal
Open MIND
Published
2026-04-15
DOI
https://doi.org/10.5281/zenodo.19600171
Primary Topic
Multi-Agent Systems and Negotiation
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

Polycode v1: A Context-Window-Free Agentic Coding CLI

Andrew Salvo
Open MIND
Multi-Agent Systems and Negotiation
article

Polycode v1: A Context-Window-Free Agentic Coding CLI

Andrew Salvo
article en

Abstract

This is the release report for polycode v1, an open-source agentic coding command-line interface built by Polylogic AI. It describes what polycode is, where it came from, what it demonstrates, what it does not, and the v1.1.x release path from the initial v1.0.0 cut to v1.1.5. The companion academic paper, Compile: A Fourth Primitive for Canon-Keeping AI Agents (Salvo 2026d), formalizes the theoretical contribution and reports the fifteen-test validation harness. This report and the paper should be read together. Polycode operates without holding conversation state in the language model's context window. State lives in a SHA-256-chained, user-owned, append-only canon on the developer's local machine at ~/.polycode/canon/. The prompt passed to the language model on each turn is a structural query of that canon, bounded in size irrespective of how much history the canon has accumulated. This applies the type-separation architecture published in Engine, Rules, and Canon (Salvo 2026b) to the developer CLI surface for the first time, and in doing so, relaxes the coupling between agent coherence and context-window size that the current generation of agentic tools relies on. Every turn in polycode mints a witnessed commitment: a canon row containing the user's claim, the tool substrate it acted on, and a verdict produced by a deterministic non-LLM composer over the witness primitives. The composer's non-LLM property is the structural safeguard the paper proved necessary at n=30 in Salvo (2026c), where nine transformer reviewers from disjoint provider families agreed at Cohen's κ=1.000 on thirty NeurIPS 2024 papers, while the structurally non-transformer witness arm refused to commit on all thirty. Substrate, the paper concluded, cannot witness itself. Install: npx @polylogicai/polycode@latest on macOS, Linux, or Windows. Zero configuration, free hosted inference tier. Source: github.com/polylogicai/polycode (MIT) Package: npmjs.com/package/@polylogicai/polycode Disclosure. Polycode is inspired by Claude Code's public architecture at docs.claude.com. Polycode is not affiliated with Anthropic. No Claude Code code or system prompts are copied into polycode. Anthropic (compile tier) and Groq (generator tier) are consumed as arms-length paid API providers; neither has reviewed or endorsed this report. Chain: Derived from Salvo 2026b, continues Salvo 2026c, companion to the Compile paper (Salvo 2026d), supplemented by the polycode and polybrain-kernel repositories. PDF is Bitcoin-anchored via OpenTimestamps. License: CC BY 4.0 (this document). MIT (polycode source). Axiom: substrate cannot witness itself.

Open MIND
Pennsylvania State University (US)
Peace, Justice and strong institutions
Openalex Percentile: Top 6%
Multi-Agent Systems and Negotiation
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.