A Wiki for Your Coding Agents
AGENTS.md for your whole org, with a human-readable audit trail and agent tooling on top. Stop losing institutional knowledge between sessions.
THE PROBLEM
Coding agents are stateless. Every session starts fresh. When an agent discovers that your team has a shared logging library, fixes the issue, and moves on — that knowledge dies.
AGENTS.md / agent.md
Single-project scope. No cross-project knowledge. Manually maintained by humans.
LIMITED SCOPE
Confluence / Notion
Built for humans, not queryable by agents. Knowledge exists but agents can't reach it.
WRONG AUDIENCE
RAG Over Codebase
Finds what is — not what should be. Can't capture conventions, decisions, or lessons learned.
WRONG DATATHE SOLUTION
vibedecoding.io is a persistent organizational memory layer for coding agents — a wiki that agents can read from and write to, continuously enriched by both humans and automated agents.
The Human Layer
- Markdown wiki, navigable via a web portal
- Humans browse, edit, and review like any wiki
- Plain markdown is the canonical source of truth
The Agent Layer
- Structured, indexed representation optimized for agent queries
- Rebuilt automatically whenever the wiki changes — on write, on push
- No manual compile step; index maintenance is invisible to users
- No magic black box — the source is always auditable
Git-Backed Storage
The wiki is a git repository. Full provenance for free — every fact has an author and timestamp. PR-based review flow for agent-proposed changes. Familiar editing workflow for humans.
AGENTS.md vs VIBEDECODING
They complement each other. vibedecoding can inject a relevant summary into AGENTS.md at project init.
| AGENTS.md / agent.md | vibedecoding | |
|---|---|---|
| Scope | Single project | Org, project, and user layers |
| Author | Human, manually | Human + agent-assisted |
| Location | In the repo | External, centralized |
| Lifespan | Lives with the repo | Persistent, independent |
| Example | "Run tests with npm test" | "Always use lib/logger.ts for structured logging" |
THE CLI
The primary interface for agents and humans alike. Works with Claude Code, Cursor, Aider — any agent framework. No API integrations to maintain.
Get a named skill — org, project, and user layers merged into one tailored output. No keywords required.
Ad-hoc query when no skill exists yet. Get org-wide conventions relevant to your task.
Write a discovered convention back to the wiki. Indexing happens automatically.
Create a named skill. The system populates and maintains it across all layers as the wiki grows.
Agent Bootstrap
The only integration required — add this to your agent.md:
SKILLS
Skills are the primary interface between the wiki and agents. A skill is a named, layered, living context bundle — everything an agent needs for a specific task, assembled from multiple scopes.
Layered Scopes
Skills are composed from three layers, merged in order. When an agent calls vibe skill code-review --project voice-tools --user ian, it gets all three layers merged into one coherent output.
Layers annotate rather than override. Both facts surface with context — the agent reasons about the tension.
You Name the Intent
Create a skill and the system populates and maintains it across all layers as the wiki grows. New facts are automatically routed to relevant skills.
The System Surfaces Patterns
Written 8 facts about auth across 3 projects? The system suggests an auth skill. Accept, reject, or rename — then it's maintained going forward.
AGENTS
There are three types of agents. Each has a distinct job.
YOU OWN
The prompts, configuration, and trigger rules. Define what agents do and when they run.
WE PROVIDE
Agent code, VM runtime, tool integrations, and execution guarantees. Reliable, isolated execution.
Event Sources → Agents → Wiki
Event Agents
Write new knowledge to the wiki. Triggered by things that happen in your workflow — a release ships, a milestone closes, a PR gets merged. Each agent watches for a specific event and writes a structured summary into the right place.
- Release Summarizer
- Milestone Summarizer
- Pattern Extractor
The Curator
Keeps the wiki itself healthy. Runs on a schedule — not per write. Detects duplicate facts, flags contradictions between entries, and proposes merges. It never deletes anything unilaterally. It proposes, and a human approves.
- Dedup facts
- Flag contradictions
- Propose merges
Skill Agents
Keep the query layer current. The Skill Maintainer watches for new wiki entries and routes them into the right skills automatically. The Skill Discoverer runs weekly and surfaces suggestions when patterns emerge.
- Skill Maintainer
- Skill Discoverer
Event agents grow the wiki. The curator keeps it clean. Skill agents keep it useful. Three jobs, intentionally separate.
Customer-Authored Prompts
You write the prompts. We run them in isolated Firecracker microVMs — ~125ms boot time, clean VM per run, no cross-customer contamination.
WHY VIBEDECODING
STANDALONE CLI
Works with Claude Code, Cursor, Aider, any agent framework. No API integrations to maintain.
SKILLS, NOT QUERIES
Agents call vibe skill code-review and get everything they need. Gets smarter over time without any agent changes.
LAYERED CONTEXT
Org, project, and user layers merge into one tailored skill. The same skill name returns different context depending on where you are.
SKILL DISCOVERY
The system tells you what skills you should have, based on what you've already written.
GIT-BACKED TRUST
Agent-proposed changes go through PR review. Full provenance. Human-auditable at all times.
FIRECRACKER ISOLATION
Real VM-level isolation for agent execution. Genuinely safe to give write access to a wiki that matters.
PROMPT-OWNED
You control the voice and focus of agents. We guarantee reliable, isolated execution.
CROSS-PROJECT SCOPE
The thing AGENTS.md can't do. Org-wide knowledge that spans every repo.
PRICING
Gate by run volume, not by agent type or feature access. All tiers get the same agents. Flat team tiers — no per-seat friction.
- 1 user
- 5 agent runs / month
- CLI access
- Git-backed wiki
- Up to 15 users
- 200 agent runs / month
- All agent types included
- $0.15 / overage run
- Up to 50 users
- 500 agent runs / month
- All agent types included
- $0.15 / overage run
- Unlimited users
- Custom run volume
- All agent types included
- SSO, audit logs
GET EARLY ACCESS
We're building vibedecoding in the open. Sign up to be among the first teams to try it — and help shape what we build next.