Graphify
Status: built-in plugin. Ships with Sero and is available in every profile.
Overview
Graphify builds a knowledge graph of your code across every workspace you opt in to, then merges them into one profile-wide graph your agent can search across. After the first build, updates happen automatically in the background — no rescanning, no manual triggers.
Getting started
- Open Graphify from the sidebar.
- Toggle on the workspaces you want indexed, or click Index all.
- Wait for the first build to finish — this is the only step that uses your AI provider.
- Ask the agent questions that span your projects: "What calls into the auth module?", "How does the payment flow connect to the user profile?"
The first build can take a few minutes depending on workspace size. After that, incremental updates run quietly after each coding session — no cost, no wait.
What the agent can do
Once a workspace is indexed, the agent has six new tools it uses automatically:
| Tool | What it does |
|---|---|
graphify_search |
Search across all indexed workspaces at once |
graphify_query |
Explore the current workspace's graph — broad or trace mode |
graphify_path |
Find the shortest connection between two concepts |
graphify_explain |
Get a plain-language description of how something fits in |
graphify_status |
See which workspaces are indexed and graph sizes |
graphify_index |
Enable, disable, rebuild, or refresh any workspace |
You don't need to invoke these manually — the agent picks them up when relevant.
Auto-context
Graphify quietly enriches every session in an indexed workspace:
- Session start — the agent reads a brief orientation from your workspace's graph before the first message.
- Search hints — when you run a broad search, graph context is appended to help the agent navigate.
Both behaviours are on by default and can be turned off in Graphify's settings.
Settings
| Setting | Default | Notes |
|---|---|---|
| AI backend | Your configured provider | Claude, OpenAI, Gemini, DeepSeek, Kimi, or Ollama |
| Model override | Backend default | Leave blank to use the provider's default model |
| Token budget | Unlimited | Cap this if you want to limit build cost |
| Exclude patterns | node_modules, dist, and common build artefacts |
Add any paths you don't want indexed |
| Auto-context on session start | On | Session orientation from GRAPH_REPORT.md |
| Graph hints in search results | On | Appends graph context to broad searches |
Privacy and data
- Your code is read once during the initial build, using your own provider credentials — nothing is sent to Sero.
- Routine updates (file structure and AST changes) run with no LLM calls.
- Graph files are stored locally at
~/.sero/apps/graphify/and never inside your repos. - Container sessions only need read access to the graph directory — your code stays inside its container.
If something goes wrong: check workspace toggle state in the Graphify panel, confirm your provider credentials are set, then restart Sero. Collect redacted logs only — do not include API keys, workspace paths, or private code content in support reports.