Stencil Labs, Inc
Stencil Labs builds new tools and instruments for agentic software development, aiming to make collaboration among engineers, reviewers, designers, managers, and AI agents seamless without sacrificing quality or individual judgment. It operates as a research company that experiments with, evaluates, and publishes improvements to the software-building workflow.
At a Glance
- Software engineers and developers
- Engineering organizations and enterprise development teams
- Designers, reviewers, and managers collaborating on software
- AI/coding-agent researchers and open-source contributors
AI Tools by Stencil Labs, Inc
(1)
omp (oh-my-pi)
Terminal AI Coding Agent CLI
Discussions
No discussions yet
Be the first to start a discussion about Stencil Labs, Inc
Latest News
The Harness Playbook: Stencil described omp² and a harness architecture based on durable state, trusted control, bounded work, compatibility, and projected views.
Prewalk: Stencil reported that sharing a frontier model's trajectory with a cheaper executor retained 92–97% of frontier pass rate at roughly half the cost on SWE-Bench Pro.
Snapcompact: Stencil published a near-verbatim visual context-compaction method reporting approximately one-third input price across four frontier models.
Hashline / The Harness Problem: Stencil reported that content-hash line anchors beat patch editing for 14 of 16 coding models, with zero training compute.
Products & Services
Open-source, model-agnostic coding-agent harness and terminal agent, built as Stencil's proving ground for applied AI research. It has a native Rust core and includes subagents, plan mode, LSP, DAP, hindsight memory, hashline edits, time-travel/session capabilities, and broad model/provider support.
A Rust-native workspace and neoterminal designed to work across local and remote hosts, with persistent terminal sessions, remote connections, browser/WebGPU access, and omp integration.
An edit format that tags file lines with short content hashes so agents can make verifiable line/range edits without reproducing old file contents; stale hashes cause edits to be rejected rather than silently corrupting files.
A research technique in which a cheaper model receives a frontier model's explored trajectory rather than only a short plan, retaining most frontier pass performance while reducing cost and completion time.
Market Position
Stencil positions itself as a research-led, open-source and model-agnostic alternative to proprietary coding-agent workflows, treating the harness—tools, state, editing, context, and interface—as a first-class product rather than assuming the model alone is the differentiator. Its own research contrasts harness approaches with Claude Code, OpenCode, Pi, Codex, Cursor, and Aider, while omp combines a native Rust core, IDE/developer tooling, multiple model providers, and experimental techniques such as Hashline, Prewalk, and Snapcompact.
Leadership
Founders
Can Bölük
Founder and CEO; systems engineer and security researcher with a decade spent across kernels, reverse engineering, and compilers. Leads product direction and R&D.
Meehir Patel
Co-founder and COO; engineering leader and technical director with years of experience running engineering organizations. Leads partnerships and enterprise adoption from San Francisco.
Executive Team
Can Bölük
Founder & CEO
Systems engineer and security researcher with experience across kernels, reverse engineering, and compilers; leads product direction and R&D.
Meehir Patel
Co-founder & COO
Engineering leader and technical director with experience running engineering organizations; leads partnerships and enterprise adoption.
Founding Story
Stencil says that current generative-model workflows still involve substantial friction even for domain experts. It was started to overcome that friction by building a collaborative 'whiteboard' for agentic development and by using practical tools as a proving ground for continuous research, evaluation, and iteration.
Business Model
Revenue Model
Stencil's legal terms describe paid subscriptions and hosted inference, with subscription-included usage credits, additional purchased credits, and usage-based charges where permitted. The company also publishes open-source software such as omp under applicable open-source licenses.
Pricing Tiers
Recurring subscriptions may include a per-period allotment of hosted-inference credits; current pricing is presented in the Services before purchase.
Hosted inference and usage beyond included or purchased credits are charged at published rates.
Target Markets
- Software engineers and developers
- Engineering organizations and enterprise development teams
- Designers, reviewers, and managers collaborating on software
- AI/coding-agent researchers and open-source contributors
- Interactive AI-assisted software development in a terminal
- Long-horizon coding tasks requiring durable, rewindable, or forkable sessions
- Multi-agent and subagent software workflows
- Remote development and driving agents from another device
- Automated software factories operating against untrusted repositories or inputs
- Research and benchmarking of coding models, edit formats, context handling, and agent harnesses