Apache-2.0 tool that builds a local code graph and evidence-citing Wiki, exposed to DeepSeek Harness as a bundle with `explore_context` and `dependency_subgraph`.
DSH integration
Native runtime
Author-claimed
Safety audit
Unaudited
Last verified
2026-11-12
License
Apache-2.0
01What can it help you accomplish?
Give the DSH agent compiler-resolved navigation of a local checkout — definitions, callers, dependencies — instead of grep-and-read
A local search index plus a typed symbol graph exposed as tools such as `explore_context` and `dependency_subgraph`, with ranked source anchors
DSH users working in unfamiliar or large repositories who need the agent to cite files and lines
Read how an unfamiliar repository works before touching it
A generated Wiki where each page cites a file and line range, the Overview opens on one recorded call path, and the map is built from the recorded references
Developers onboarding onto a codebase, or reviewers who want an evidence-backed map instead of model prose
02How to install into DeepSeek Harness
Prerequisites
- Python 3.10+ and Git — the README lists both for the code-graph setup
- Language toolchains for the languages you index; the README says registered capabilities are not the same as the tools installed on your machine and points at `codenib toolchain status`
- No model, API key or GPU is needed for the code graph itself — the host agent uses its own model
Installation steps
- 01
Install the DSH bundle from the repository's `plugins/dsh` path: `dsh plugin add "git+https://github.com/sysevol-ai/CodeNib.git#main&path:/plugins/dsh"`
$ dsh plugin add "git+https://github.com/sysevol-ai/CodeNib.git#main&path:/plugins/dsh"
- 02
Install the Python side with `python -m pip install "codenib[graph,mcp]"` and initialize the index with `codenib codegraph init /path/to/your/repo` — initialization accepts uncommitted edits and preserves the checkout
- 03
Ask the agent something like “Use CodeNib's `explore_context` to find where request retry behavior is implemented. Cite the files and lines.”
Verify the integration
- `codenib toolchain status` reports which language backends are actually available on the machine
03DSH integration and capability boundaries
DeepSeek Harness bundle installed from the repository's `plugins/dsh` path, giving the host agent CodeNib's local code-graph tools
Local code graph for the host agent
A repository checkout, including uncommitted edits→A local search index and typed symbol graph that the installed agent clients can query
The index is built locally; the checkout itself is preserved`explore_context` and `dependency_subgraph`
A plain-language question, or a symbol to expand→Ranked source anchors for a question, and incoming/outgoing static references in one call with depth ≤ 8
Indexed references can include calls, fields, type uses and function values, so which reference actually invokes a function has to be confirmed in the sourceWiki generation
`codenib wiki /path/to/your/repo --preset graph`→A local Wiki at `http://localhost:3000` with one page per code area, each citing a file and line range, plus a map of recorded references
Pages are written from the index without a model; `--generate --model <provider/model>` switches to model-written pagesRegistered language support
Repositories in the supported languages→A code graph for 12 of the 14 registered languages, with an incremental LSP or clangd backend for Python, Go, Rust, C/C++, JavaScript and TypeScript
Swift and Lua are search-only, without definitions or callers; the registered list is not a statement about installed toolchains
04Who is it for? When not to use it?
Good for
- DSH users working in unfamiliar or large repositories who need the agent to cite files and lines
- Developers onboarding onto a codebase, or reviewers who want an evidence-backed map instead of model prose
Not for
- The indexed graph can miss external and dynamic relationships, and a reference can be a call, field access, type use or function value — you still have to inspect the source to know which one invokes a function.
- CodeNib is in beta and public interfaces may change before a stable release; registered language capabilities are not a guarantee that the needed toolchains are installed.
- The optional grep → Jev retrieval route plans and reranks with OpenRouter, so selected code is sent to remote models; it is separate from the model-free code graph.
05Compatibility, maintenance and safety notes
- The indexed graph can miss external and dynamic relationships, and a reference can be a call, field access, type use or function value — you still have to inspect the source to know which one invokes a function.
- CodeNib is in beta and public interfaces may change before a stable release; registered language capabilities are not a guarantee that the needed toolchains are installed.
- The optional grep → Jev retrieval route plans and reranks with OpenRouter, so selected code is sent to remote models; it is separate from the model-free code graph.
- The ContextBench leaderboard result describes CodeNib's research agent (Haiku workers plus an Opus repair step), not the Wiki or the code graph that this plugin exposes.
Apache-2.0 source; every claim here comes from the repository README at the recorded commit, with no independent safety audit.
06Frequently asked questions
How is CodeNib installed into DeepSeek Harness?
The README installs it as a DSH bundle: `dsh plugin add "git+https://github.com/sysevol-ai/CodeNib.git#main&path:/plugins/dsh"`. The Python side is `python -m pip install "codenib[graph,mcp]"`, followed by `codenib codegraph init /path/to/your/repo`.
Does the agent need its own API key or model for the code graph?
No. CodeNib builds local search and a typed symbol graph with SCIP/LSP indexers and needs no model, API key or GPU of its own; your agent uses its own model. Only the optional grep → Jev route calls remote models through OpenRouter.
What can the DSH agent actually ask?
`explore_context` answers a plain-language question with ranked source anchors, and one `dependency_subgraph` call returns incoming and outgoing static references down to depth 8. The README cautions that references may include calls, fields, type uses and function values, so the source still decides which one invokes a function.
Which languages get a real code graph?
Twelve of the 14 registered languages: Python, Go, Rust, C/C++, JavaScript and TypeScript get an incremental LSP or clangd backend; C#, Java, Ruby, PHP, Kotlin and Scala get a code graph; Swift and Lua are search-only. `codenib toolchain status` shows what is actually installed.
Are the published benchmark numbers about the Wiki or the plugin?
No. The README says the ContextBench leaderboard result (77.90% line recall, 77.60% Pass@1) is CodeNib's research agent — two Haiku workers searching plus an Opus repair step — not the Wiki or the code graph this plugin exposes.
07Related DSH workflows
dsh-our-free-model
by ebony-vinyl
Adds model choices, availability checks, usage panels, and a local OpenAI-compatible forwarder.
dsh-routing-suite
by yjh051108
dsh-routing-suite — injector + router-standard kit: install the runtime injector first, then the task-aware reasoning-mode router preset (measured P1-P23).
dsh-ops
by t-auto
Run Bash and PowerShell commands, search and edit local files, and manage long-running background jobs from DSH.
brooks-lint
by hyhmrright
Portable AI code quality reviews for Codex, Cursor, Copilot and other coding agents: six Agent Skills for architecture, technical debt, tests and remediation.
08Data and sources
Grok Build reads the same plugin, and DeepSeek Harness installs it as a bundle:
dsh plugin add "git+https://github.com/sysevol-ai/CodeNib.git#main&path:/plugins/dsh"
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-11-12. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.