First inspect the pinned package, patch, README, and linked security documentation. Confirm that your DSH Web profile and Node version satisfy the declared prerequisites. Because the README command uses @latest, this page intentionally does not repeat it as a pinned installer; use the upstream material to choose a release after evaluating its provenance.
BundleVision / Model integrationv3.14.0Developer Preview
ModLens
ModLens is a DSH Web bundle that routes pasted-image evidence for text-only DeepSeek and GLM model variants through a configurable vision engine.
1,029 repository starsMIT Updated Aug 14, 2026
Review snapshot
- Manifest
- Passed
- Install
- Not tested
- Runtime
- Not tested
Static compatibility snapshot dated 2026-08-14; install and runtime remain independent checks.
Install / upstream documentation
Install the pinned release deliberately
This bundle uses an upstream-documented installation path pinned to the reviewed release. Read the linked README and inspect the source before you install it.
Upstream command · profile web
npx -y @deepseek-ai/dsh plugin --profile web add @liustack/modlens@3.14.0The pinned README documents this DSH installation shape with @latest. This listing replaces latest with the reviewed npm release 3.14.0; its npm gitHead differs from this repository snapshot, so inspect both sources before installing.
Source: Pinned upstream DSH installation notePlugin brief
What ModLens adds to DeepSeek Harness
ModLens is a DeepSeek Harness plugin packaged as @liustack/modlens. Its pinned package.json declares a DSH bundle patch and a Web client; the pinned cordis.patch.yml inserts the modlens row. The README describes its purpose as giving supported text-only DeepSeek and GLM model variants a way to obtain image evidence, including images pasted directly into a chat. This page records the source snapshot, not an independent claim that every model or DSH release will behave that way.
The design is a bridge rather than a new foundation model. Upstream says users choose the DeepSeek-V4-Flash (modlens vision) or DeepSeek-V4-Pro (modlens vision) entries for pasted images, while the bridge excludes native vision variants. It then relies on a configured vision engine to produce evidence that the text-only route can use. That division matters: image content, provider availability, latency, and any account quota are outside the normal text-only model path.
The README describes several provider choices: direct Gemini, OpenAI-compatible endpoints, Anthropic, Antigravity CLI, Claude CLI, and opt-in reuse of other local agent CLIs. It presents failover as an explicit sequence and says reused reads are labeled with the quota source. Those are upstream descriptions, not a guarantee that a particular provider is free, permitted, private, or compatible with your organization. Read the upstream security and configuration documentation before attaching accounts or API keys.
ModLens can be attractive for screenshot questions, diagrams, slides, charts, or UI debugging in a DSH conversation. It should not turn a screenshot into unreviewed truth. Vision output can omit details or make incorrect inferences, while the pasted image itself may contain hostile instructions. Keep the original file available, verify consequential facts manually, and avoid sending confidential images to a provider until you understand where the data will travel.
Evaluation workflow
Evaluate ModLens with clear boundaries
Choose one vision-provider path deliberately. A direct API may require a key and sends image data to that service; CLI reuse may draw on an existing subscription or login. Do not enable reuse merely because a local CLI is detected. Record which account is allowed to handle images and what quota or billing policy applies.
Test with a non-sensitive image and a question with an answer you can independently verify. Check the selected DSH model variant, whether the image remains represented as expected in the conversation, which provider handled the request, and whether the result distinguishes observed text from inference. Treat a health check as an operational signal, not as a privacy or accuracy approval.
For recurring use, constrain configuration ownership and periodically review provider settings, proxy configuration, and local CLI permissions. The repository describes remote URL and image-handling security considerations; an image may embed untrusted content just as a web page can. Keep secrets out of prompts and screenshots, and revoke or rotate credentials through the provider if exposure is suspected.
Verification ledger
Evidence, not installation claims
Manifest
PassedPinned package.json declares dsh.bundle.patch as ./cordis.patch.yml; the pinned YAML patch inserts id modlens with name @liustack/modlens.
Checked Aug 14, 2026
Install
Not testedNo clean DSH Web installation was run.
Runtime
Not testedNo provider-backed pasted-image test was run.
| Layer | Result | What was checked |
|---|---|---|
| Manifest | Passed | Pinned package.json declares dsh.bundle.patch as ./cordis.patch.yml; the pinned YAML patch inserts id modlens with name @liustack/modlens.[Aug 14, 2026] |
| Install | Not tested | No clean DSH Web installation was run. |
| Runtime | Not tested | No provider-backed pasted-image test was run. |
Source or manifest review can be recorded while Install and Runtime remain not-tested. Treat these as separate verification layers.
Execution surface
Prepare, build, and code-execution review
prepare
not recorded
No review record is available for this execution surface. Inspect the upstream repository before running it.
build
not recorded
No review record is available for this execution surface. Inspect the upstream repository before running it.
code execution
review recorded
The package exposes a CLI and DSH integration. Review the pinned source and any installer output before execution.
network
review recorded
Vision providers and reused CLIs can send image evidence and prompts to remote services; review each provider’s data handling and terms.
configuration
review recorded
Provider keys, proxy settings, and opt-in reuse of existing CLI logins determine where requests run and whose quota may be used.
Third-party code can execute during dependency preparation, build, or runtime. Read the upstream source and its changes before use.
Declared scope
Capabilities and prerequisites
Capabilities
- Adds a DSH Web bundle row named modlens
- Documents read_image support for pasted images on declared DeepSeek and GLM text-model variants
- Supports configurable vision providers and a failover chain
- Provides a CLI, health-oriented documentation, and configuration references
Prerequisites
- DeepSeek Harness Web profile
- Node.js 22.19 or newer, as declared in package.json
- A configured or explicitly reused vision provider/CLI account
- Careful review of image, provider, and account-data handling
Frequently asked questions
ModLens FAQ
What is ModLens?
ModLens is a DSH Web bundle that upstream describes as a vision bridge for selected text-only DeepSeek and GLM model variants, including pasted-image workflows.
Is ModLens a real DSH Bundle?
Yes, static evidence supports that classification: package.json declares dsh.bundle.patch and the pinned cordis patch inserts @liustack/modlens as modlens.
Why does this page pin ModLens 3.14.0?
The pinned README uses @latest, while this page replaces that mutable tag with the reviewed npm release 3.14.0. The npm gitHead differs from the repository snapshot, so inspect both sources before installing.
Which vision services can it use?
The pinned README describes direct providers and opt-in reuse of local agent CLIs. Availability, pricing, terms, and privacy are determined by those providers and your configuration.
Can I paste confidential images?
Do not do so until you have reviewed the selected provider, network path, organizational policy, and configuration. A vision request can transmit image content outside DSH.
Has AgentPlugin.net tested ModLens?
No. Manifest and patch evidence were inspected statically; installation, provider configuration, pasted-image behavior, and runtime remain untested.