The honest answers.

It is a memory layer for AI coding agents. It ties your code, tickets, docs and threads into one graph, then serves that graph to whatever agent you use, so the agent stops starting every session from zero.

Documentation explains what happened. Memory explains why, and stays true as the code changes. A wiki is written by hand and drifts; Oynix forms from the work already happening and is linked to the code, so it cannot silently go stale.

No. Vector search returns chunks of text that look similar to your query. Oynix returns the exact nodes and the paths between them: who calls this, who owns it, why it exists. It uses embeddings as one signal, but the graph is what sits on top.

No. Oynix has no model of its own on the retrieval path. Your agent does the reasoning. Oynix gives it the organizational context it was missing, over MCP.

Oynix runs as an MCP server. Any MCP-capable client (Claude, Cursor, Codex and others) connects to it and calls purpose-built tools: ownership, source, context, impact, blast radius. One command writes the right config for each host.

Eighteen sources today, including GitHub, GitLab, Bitbucket, Slack, Microsoft Teams, Jira, Linear, Sentry, Confluence, Notion, Google Drive, Microsoft 365, Mixpanel, CleverTap, Firebase and Supabase. Most connect in one tap; the rest take a token or your own OAuth app. You pick exactly which channels, spaces, projects or folders sync.

No. There is no charge per file, per function, per repository or per commit. A four million line monorepo costs the same as a side project. Only the number of people changes the bill.

No. Sync is incremental, with modified-time cursors, high-water marks and content-hash dedup. Re-syncing picks up what changed and skips what did not.

Local runs entirely on your machine. No account, no server, and it connects GitHub, Jira, Confluence and Google Drive. Local plus Server adds the engine a team shares: every connector, better embeddings and reasoning, Oynix Ask, Oynix Wiki, and one memory everybody reads from.

Yours. Oynix ships the engine; it runs on your infrastructure, on your database. We never host your code or your graph. The server in Local plus Server is your server.

Yes. A graph nobody shares has nothing to bill for. Local makes one engineer faster and costs nothing. The business starts the moment a second person needs the same answer and should not have to go and ask for it.

No. Retrieval runs as an MCP server, so your agent does the reasoning and we return graph facts. There is no model call on our side to bill you for. Your AI provider bills you for its own usage, on your own key.

On Local plus Server: Solo is 15 dollars a month for one key, Team is 99 dollars a month for 10 keys or 459 dollars a month for 50, and Enterprise is custom. Every paid plan includes all 18 connectors, better reasoning, Oynix Ask and Oynix Wiki.

It stays. Downgrading suspends collaboration and the extra seats. It never deletes what Oynix already learned. Losing your memory is the exact problem this was built to end, so we do not do it to you over a billing cycle.

A key is a rotatable credential bound to a person, not an account. It carries that person's role. If a key leaks, you rotate it without touching the user, the workspace or the data behind it.

In your database, on your infrastructure. Oynix ships the engine, not the vault. We hold key metadata so a leaked credential can be rotated. We never hold your code, your graph, or your provider keys.

No. The indexing engine runs inside your own infrastructure and writes to a database you control. Your source never reaches us. Most tools ask you to upload your code to earn their answers. Oynix does the opposite.

Bring your own. They live on your engine and never touch our servers. When you rotate them, nothing on our side needs to change.

Roles are Master, Admin and Member. Billing and workspace deletion sit with Master, configuration and integrations with Admin, and Members query and use MCP. A workspace always keeps at least one Master.

Install locally with one command and point it at a repo. Nothing to sign up for. When your team is ready to share one memory, add the server.

Any MCP client. That includes Claude Code, Cursor, VS Code, Zed, Gemini CLI, Codex, Cline, Roo Code, OpenCode, OpenClaw, Devin, Kiro, Trae and Kimi CLI. Eighteen hosts and counting.

Yes. The graph is namespaced across every repo, so cross-repo ownership, callers and blast radius resolve as one structure rather than per-project silos.