Knowhere API - Transform Documents into Structured Data

Your docs

.pdf.docx.xlsx.ppt

become

OpenClaw-nativecontext

with grounded retrieval

The plugin uses Knowhere for parsing and job orchestration, stores the returned result package inside OpenClaw-managed local storage, and gives agents a browse-first path to previews, chunks, hierarchy, and raw files before they answer.

See integration guide Get API key

{ Browse-first }

{ Path-aware }

{ Chunk-backed }

{ Citation-ready }

{ OpenClaw-native }

Result packages

Store once. Reopen anytime.

Knowhere parses the document, OpenClaw stores the returned package locally, and agents can reopen the exact manifest, hierarchy, chunks, and raw files later.

Tool surface

Browse before the answer.

The plugin registers `knowhere_*` tools for preview, grep, raw-file reads, cleanup, and explicit ingest flows instead of forcing everything through one opaque call.

Auto-grounding

Context arrives when it matters.

When `autoGrounding` is enabled, OpenClaw can auto-ingest attachments and inject compact document availability or status context right into the agent loop.

Grounded Answer Flow

One dense report in. One grounded OpenClaw answer out.

This is the interaction model the plugin is built for: Knowhere extracts structure, OpenClaw stores the package, and the agent answers only after it has previewed or reopened the right evidence.

UNSTRUCTURED

Raw source:

Tesla-Q4-2025-Update.pdf

STRUCTURED

OpenClaw

knowhere skill loaded

Did Tesla's free cash flow go negative in any quarter? Show the supporting chunk.

Yes. Q1 2024 is the only negative quarter. Operating cash fell to $242M while CapEx stayed at $2,777M.

−$2,535M

manifest.jsonchunks.jsonpage-33 / table-14

What should I inspect if I want the raw source instead of the answer?

Open the preview first, grep for the metric, then read the exact result file behind that chunk. The plugin keeps the path surface intact.

preview → grep → read_result_file

knowhere_preview_documentknowhere_grepknowhere_read_result_file

Integration Guide

Install it in OpenClaw in three commands.

Follow the same rhythm as a developer-tool homepage: read the steps once, copy the commands in order, and replace the API key only in step 02.

Package

@ontos-ai/knowhere-claw

No config wall, no runtime internals, and no extra surface to learn. Install the package, attach the API key, then enable the plugin.

ClawHub Skill

Knowhere

If you install from ClawHub, look for the skill named Knowhere.

1. Install package

Add the packaged runtime so OpenClaw can load the bundled knowhere skill.

$ openclaw plugins install @ontos-ai/knowhere-claw

2. Attach API key

Connect this OpenClaw instance to your Knowhere account.

Everything else can be pasted exactly as shown. The API key line is the only place where you replace a value.

$ openclaw config set plugins .entries.knowhere.config.apiKey "sk_..."

3. Enable plugin

Turn the entry on so agents can load the plugin inside the runtime.

$ openclaw plugins enable knowhere

Bringknowhereinto

OpenClaw.

Install the plugin, point it at your API key, and give OpenClaw a browse-first way to inspect documents before an agent answers.

Get API key Review install steps

PDFs, scanned files, tables, manifests, chunks, and raw result files stay reopenable instead of disappearing into one generated reply.

What changes inside OpenClaw

knowhere_* tools

Preview, grep, raw-file reads, ingest, and cleanup become callable in one place.

Browse-first evidence

Agents can reopen manifest, hierarchy, chunks, and raw files before answering.

Scoped local storage

Result packages stay reusable across session, agent, or global scopes.