Repository navigation
Releases
Releases · asta-nguyen/openez-graph
Version 1.7.1
[1.7.1] - 2026-11-07
Run openez prune after upgrading to list workspaces that earlier versions left in the registry.
Added
openez prunelists registry entries whose directory no longer exists (deleted worktrees, temp dirs) and, with-y, removes them. A/entry left by an earlier version is included.
Changed
- MCP startup registers and indexes its working directory only when it is inside a git repository, and registers the repository root instead of a subdirectory. Hosts that launch the server from
/,~/.codex,$HOMEor a temp dir no longer create workspaces there; runopenez init <path>to index such a directory. A repository rooted at$HOMEis skipped.
Fixed
- A registered workspace with root
/no longer breaks/api/memoriesand/api/documentswith "Workspace root path does not exist". The dashboard, documents, memories and metrics now choose their workspace through one check that skips an unusable root. openez init,index,watchand the MCP server refuse the filesystem root as a workspace root instead of registering it.
Full Changelog: v1.7.0...v1.7.1
Version 1.7.0
[1.7.0] - 2026-11-06
Run openez reindex <path> after upgrading: symbol extraction changed (oxc-v6).
Added
.openezignore: a root file with.gitignoresyntax for paths that stay in Git but should not be indexed.openez initandopenez setup <agent>create one with common patterns when the project has none, and never overwrite an existing file.code_queryranks files whose name the question spells out, so "React Query client" findslib/query/queryClient.ts.code_queryranks the file defining an exact symbol name first, including members found by bare name (connect_or_spawnfindsDaemonClient::connect_or_spawn).code_queryanswers "what calls X" and "what does X call" from the call graph, including calls from functions nested inside X.openez setup claudeadds a project hook that reminds the agent once per session to trycode_querybefore grepping.openez setup codexandopenez setup devinadd the same reminder at session start; Codex runs it only after you trust it with/hooks. None of them block a tool call.- Truncated MCP responses report
metrics.fullResponseTokens, themaxTokensvalue that would return the whole result. - A comparison with Graphify on the benchmark page and in
BENCHMARK.md.
Changed
- After graph expansion, the best match, files defining a queried symbol, and documents keep their rank; graph neighbors reorder the remaining results.
- Graph neighbors are ordered by fewest connections first, so widely used utilities no longer take the limited slots.
- FTS queries match plain word stems: "truncated" finds
truncateand "chunks" findschunk. Generated*.gen.*files rank lower. - Retrieval and chunking always use the built-in defaults, because per-project configuration was removed.
Removed
brain.config.*is no longer read. Use.openezignoreto skip files. Indexing prints a warning if the file is still present.
Fixed
- A
!line in.gitignore(such as!.env.example) no longer makes a workspace index 0 files. Re-include lines are now ignored, so a file they re-include stays skipped. - Components wrapped in a call, such as
const Row = memo(function Row() {}), are now symbols; every call inside them was previously lost from the graph.
Security
- Indexing no longer executes code:
brain.config.*was evaluated when loaded, so indexing or opening an agent in an untrusted repository could run code from it. - The web UI API rejects requests addressed to a non-loopback host (DNS rebinding) and writes from a non-loopback
Origin(CSRF), which any open web page could previously send. openez setupno longer changes agent config files to mode 644, which could expose tokens stored in them. New files are created as 600 and existing files keep their mode.- JSON, YAML, and TOML files named like credentials or secrets (
*credentials*,*secret*,service-account*.json) are never indexed, so their contents cannot reach an agent.
Full Changelog: v1.6.1...v1.7.0
Version 1.6.0
[1.6.0] - 2026-10-22
Added
- Native Rust parser and file scanner binaries for Linux x64 GNU, macOS arm64, and Windows x64 MSVC, with tree-sitter and regex fallbacks on other platforms.
- Structured symbol extraction and workspace-aware call resolution for C++, C#, and Java.
- Binding-aware call graph resolution for TypeScript/JavaScript, Python, Go, Rust, and Ruby.
Changed
- Removed symbol and call graph caps for JavaScript/TypeScript, Python, Go, Rust, and Ruby to retain larger code graphs.
- Run
openez reindex <path>after upgrading to rebuild parser rows and graph artifacts for updated language indexers.
Fixed
- Improved language-specific call resolution for imports, re-exports, namespaces, module paths, lexical scopes, and local shadowing; ambiguous and unresolved calls no longer create misleading edges.
- Reduced memory churn and fitting time for large MCP
diff_contextresponses by avoiding tokenizer cache-hit LRU refreshes and batching structured trimming while retaining exact cl100k token budgets.