The layer above your agents.
AI agents work fast and tireless. Each one sees only its own slice of the project. Oystro sits above them all, holding one memory, one process, one view, so the project keeps moving and the work ships, without locking you into a single tool.
Your process. Every agent.
Every agent vendor wants you locked into their tool: their agent, their model, their workflow. Oystro is agent-agnostic. The workflow is the constant. The agent is interchangeable.
The agent agrees with you, and neither of you notices.
Decisions happen in the prompt, invisible and unaudited. The agent fills missing context with plausible guesses that look right. Weeks later, the whole feature is built on a wrong assumption, and you only find out when it ships.
- ✕No project memory across sessions
- ✕No gates, no moment to stop and check
- ✕Confident wrong answers, at speed
The agent stops at every human gate.
You review the design before a line of code is written. You approve the contract. Then the agent executes, one task at a time, no improvisation, every decision recorded. What you approve is what ships.
- ✓Durable memory across every session
- ✓Verification gates with real evidence
- ✓Traceable work you can audit and trust
The protocol is live. Start today.
No need to wait for the paid launch to get value. Oystro OSS is the protocol itself, open source and running now: the workflow, templates, and rules. File-based and agent-agnostic, with no runtime, no account, and no API keys. The paid engine, Oystro Standard, adds the memory and the enforced gates that make it dependable.
The workflow
/h:init → /h:define → [Gate 1: /h:approve]
→ /h:build → [/h:review-pre-verify · L only]
→ /h:verify → [Gate 2: /h:release]Try it. Two minutes.
read https://github.com/oystro/oystro-oss/blob/main/commands/init.md
# and follow the instructions to initialize oystro-oss in this folderThe scaffold appears: constitution, PRD, architecture. Then start defining features. Works with any agent that can read and write files.
The control plane for AI-driven development.
You don't have to trust any single agent. You trust the layer above them all. The protocol runs free today as Oystro OSS; the engine that enforces and remembers it is Oystro Standard.
Durable project memory
Agents forget everything between sessions. The engine keeps decisions, evidence, and progress, so any session resumes in seconds, with any agent.
Verification gates & evidence
The protocol defines the gates; the human is the enforcement point. The engine makes them deterministic and captures the evidence. The gate is real: it can block.
Right model, right job
Oystro Switch routes routine work to a fast, low-cost provider and reserves premium models for architecture and hard problems.
Agent-agnostic by design
Claude, Cursor, Codex, Copilot, or whatever comes next. Bring any agent. Keep your standards. Swap agents; the workflow stays.
Resume in minutes
Multi-week projects never lose the thread. A compact handover brings a new session back up to speed, for roughly 85% fewer tokens than a full replay.
Traceability without ceremony
Every commit tells the story of what was built and why. Six months later, git log tells you exactly what happened, no guessing.
The protocol is free. Oystro Standard is the engine.
Oystro Standard is the paid engine for one engineer: command-line tools, project memory, a code map, and enforced gates, plus a local dashboard (Oystro Visualizer). This is what we are shipping next.
Enforced gates
The protocol defines the gates; the engine enforces them. Prerequisite checks, approval state, no skipped steps. The gate can actually block.
Evidence you can check
Every "done" carries proof tied to the exact commit. Fail closed: no evidence means failed, never a silent pass.
Oystro Graph
Know what a change will touch before you make it. A living map of the project, so refactoring stops being a gamble.
Oystro Search
Find code, plans, decisions, and evidence by meaning, not just exact names. One place to look, always current.
Oystro Vec
The project's memory. Ask why something was built a certain way and get the real decision, months or years later.
Oystro Switch
The right model for the job. Spend on the hard parts, not on busywork, without quietly lowering quality.
Common questions.
What is Oystro?
Oystro adds human approval gates and project memory to AI coding agents. Agents work autonomously, but they stop at two points for a person to approve: the plan before any code is written, and the release before it ships. Every decision and every test result is recorded with the project.
Is Oystro free?
The protocol itself is free and open source (Oystro OSS): the workflow, templates, rules, and command contracts, file-based with no runtime and no account. The engine that enforces the gates and keeps the memory (Oystro Standard) is paid.
Which AI agents does Oystro work with?
Any agent that can read and write files: Claude, Codex, Cursor, Copilot, OpenCode, Gemini, or whatever comes next. The workflow is the constant; the agent is interchangeable.
How is Oystro different from vibe coding?
In vibe coding, the agent generates and you accept or reject after the fact. In Oystro, the agent stops before it writes code: you approve the plan first, then review the evidence before release. The difference is who is in control.
Will Oystro make my project more maintainable?
Yes, by keeping the reasoning with the code. Every change is one commit tied to an approved plan, and the build narrative records what changed, why, and what proved it. Months later, you can still explain why the code works the way it does.
What is the difference between Oystro OSS and Oystro Standard?
Oystro OSS is the protocol reference: the method, as files you copy into your project. Oystro Standard is the engine: enforced gates, evidence capture, project memory and a code map, and a local dashboard. Start free with the protocol; add the engine when you want it enforced.
Oystro Standard launches soon.
Join the waitlist and we'll email you the moment the engine goes live. Founding users get first access and a say in what we build next.