The guide
Everything Oynix does,
in three steps.
Install it, connect your code and the tools around it, then ask your coding agent. Below is exactly what each step looks like, and every Oynix tool at work inside the agent you already use.
01 · Install
One command. No account, on your device.
One line installs a signed binary, wires it to your coding agent, and starts a local engine on your machine. Your code is indexed on your machine and nothing is uploaded.
priyam-oynix ~ % curl -fsSL https://oynix.dev/install-local.sh | sh → Downloading Oynix (darwin-arm64)… → Verifying signature… → Verifying checksum… → Checking the archive… → Checking the binaries run on this machine… → Installing to /usr/local/bin… ✓ oynix version 0.1.3 (local) installed. Next: oynix init # set up + wire your coding agent oynix connect # add repos & tools (the engine starts itself) oynix graph # see your knowledge graph The first `oynix engine start` downloads a private Python runtime (a few hundred MB). Your code is indexed on THIS machine: https://oynix.dev/security priyam-oynix ~ % oynix init ✓ Config encrypted and saved Oynix initialised! → Everything stays on this machine. Nothing is uploaded. Structural graph for now. Add an AI provider any time for search and answers. Graph stored at /Users/priyam-oynix/.oynix/local Find it later with `oynix graph path` ? Wire your coding agent (Claude Code) to Oynix now? Yes Claude Code — MCP registration ✓ oynix updated → ~/.claude.json — restart Claude Code to load it. Other clients: oynix install mcp --client claude|cursor|windsurf|gemini|cline|roo|witsy|vscode|zed|opencode|openclaw|codex|hermes|goose|kiro|qwen|antigravity|copilot|factory|trae|all ⠦ Preparing the local engine (one-time setup)…✓ Local engine started → oynix-indexer (local): http://127.0.0.1:8001 Mode: local (zero-Docker) — parser-only (no key) · graph persists in /Users/priyam-oynix/.oynix/local. Next: oynix graph see your code and your tools in one map ✓ Local engine ready Next: oynix open models add an AI provider — optional, but sources sync fully with one or: oynix connector add your folders, GitHub, Jira, Confluence, Google Drive and your own docs, all on one page priyam-oynix ~ % oynix connector → Opening the Connectors page in your browser… http://localhost:8001/connectors-ui?project=01a03981-e66e-77bc-b07f-840149144773 Connect a source there, then it syncs into your graph. Ask your coding agent about it — it can see the graph now. priyam-oynix ~ %
02 · Connect
Connect your code and the tools around it.
oynix connect opens the connectors page. Add a GitHub repo or a local folder, then bring in the docs, tickets and threads that explain it. Each source links to the code it shaped. Pick exactly what syncs.
Connect the tools your team already uses. Their tickets, docs and threads become part of the graph and link to the code they're about.
03 · Use the tools
Ask your agent. It answers with your context.
Oynix gives your agent a set of tools over MCP: read your code, size up a whole repo, and carry memory across sessions. You do not call them by hand. You ask a normal question and the agent picks the tool. Here is each one at work in Claude Code. It works the same in Cursor, Codex, Gemini and every other MCP agent.
Context
Why this exists: the tickets, threads and decisions behind it.
Answers the 'why'. It searches your org memory (Jira, Slack, Confluence, Notion and past decisions) for everything tied to a file or feature and returns the risks, the tickets, and the reasoning behind it. This is the tool that makes your agent sound like a senior teammate instead of a stranger.
- Why does this exist?
- Any known risks on the upload endpoint?
- What should I know before I change this?
When it gets usedWhen you need the why, the history a diff alone can never show you.
The real risk here is KAN-22, a race with a concurrent RunInTransaction. The standing decision is to keep the RunInTransaction wrap in apis/backup_upload.go. Do not remove it without addressing that race first.
Give your agent your team’s memory.
Free, forever, on your own machine. It takes one line.