Silicon Stencil

what: independent software studio
who: Hank, developer and architect
where: California
since: August 2025
now: hordecode · v0.1.2 · 2026-10-20 · active
contact: [email protected]
elsewhere: github · x · npm

work

hordecode2025 →An AI coding agent for the terminal. A friendly fork of opencode with its own layer: supervisor, harness and research pipelines, inbox, persistent memory, skills, workflows and cron. Your keys, more than 200 providers and local models. MIT.active
why
- Long-running work is built in.A supervisor runs delegated work as tracked child sessions with waiters and task links, so multi-step work survives past one turn. Tasks, workflows, cron and an inbox are tools the agent calls.
- Memory that stays local.Seven memory tools keep project knowledge across sessions. Keyword search works out of the box; semantic search is opt-in and runs on-device.
- A research pipeline beside the build loop.The research agent runs ask, scope, gather, verify, filter, synthesize, decide and eval. Gathering uses up to five parallel sub-agents, each with its own context; the report is written once from verified evidence and carries a citation table, and every run ends with a decision record. Built for architecture questions and R&D as much as for code.
- Skill pipelines ship with the binary.A harness set of 16 skills (plan, tdd, debug, review, verify and more) and a research set of 16 skills.
- It learns from its own sessions.Dream consolidates memory and distill packages a repeated workflow into a skill. Both run automatically on a schedule and can be turned off in config.
- Max mode.The max agent runs several reasoning candidates per step and a judge call picks the one that executes. Experimental.
- Scheduled work you approve.Cron runs jobs on a schedule; a shell command in a job runs only after you approve it.
- Everything visible.Running agents, subagents and scheduled jobs show in the sidebar with status and elapsed time.
- Your keys, your models, your machine.More than 200 providers and local models through one config. Every opencode provider, plugin and config file keeps working.
facts
Tracks the opencode 1.x line. The supervisor and harness are customized ports of the MiMo Code designs. TypeScript on Bun. Linux, macOS, Windows. Releases: 0.1.0 2026-09-13 · 0.1.1 2026-10-13 · 0.1.2 2026-10-20.
docs hordecode.com/docs · source github · changelog CHANGELOG.md · npm hordecode
phantomlogos2026A local-first agent operating system in Python that runs a hierarchy of agents on your own machine with persistent memory. Reached v1.2.1, public under MIT.paused
what it has
- Three-tier agent hierarchy.Planner, worker and critic roles with explicit hand-offs.
- Mnemosyne memory.A 14-axis persistent memory model that outlives a single run.
- Verified routing.Z3 checks the routing rules; local-hybrid routing picks local or remote models per task.
status
Paused at v1.2.1. Parts of its memory and routing work are planned to return inside hordecode.
source github
slab2025A gaming book and lore archive, the idea this studio started with. Concept art and world notes exist.paused

install hordecode

# curl -fsSL https://hordecode.com/install | bash
# npm install -g hordecode

Installs the commands horde and hordecode. opencode, if present, stays as it is.

history

2026-10-20hordecode 0.1.2Permission and skill pipeline fixes, Windows binary cleanup, config read from a directory set by environment.
2026-10-13hordecode 0.1.1, first version on npmScheduled shell commands now need approval. Memory writes validate their scope and stay inside the memory directory.
2026-09-13hordecode 0.1.0First release. opencode 1.17.18 as base, with harness, supervisor, inbox and memory ported on top.
2026-08hordecode gets its name and domainThe agent project settles on hordecode and hordecode.com. siliconstencil.com waits its turn.
2025-08studio startsSilicon Stencil is set up as the place where ideas are taken one at a time from sketch to shipped software. The first idea on the bench is slab; the agent work that follows grows into hordecode.