tesseract-ocr

DeepSeek Harness (dsh) plugin that lets text-only models accept attached images: every image is recognized locally with Tesseract OCR and only the recognized text is sent to the model API.

Privacy default: image bytes are OCR'd locally and not sent to the provider. Set passthrough: true only if you intentionally want genuine vision models to receive original image bytes.

Tested on Ubuntu (primary target); works anywhere the tesseract CLI is installed (Linux, macOS, Windows). Verified against dsh 0.1.2-rc.1 (master).

• No configuration changes to your models — no input: [text, image] hacks in settings.yaml.
• Works with any provider/model in dsh; by default every attached image is OCR'd before the request leaves the machine.
• Vision-model passthrough is opt-in (passthrough: true).
• Fail-closed: if the plugin is not loaded, models stay text-only and image attachments are refused — nothing can silently leak. Missing attachments are replaced with a refusal text block (never left as raw image).

Do not enable this plugin together with windows-ocr: both would OCR the same image. Pick one per machine.

Install from npm

dsh plugin --profile web add dsh-tesseract-ocr

(Replace web with your profile, e.g. tui.) Prebuilt and published with Sigstore provenance — no source build or allowBuilds approval needed. Installing from source (this repo) still works via the agent guide or the manual steps below.

or from the repository / a tarball:

dsh plugin --profile web add ./dsh-tesseract-ocr        # source checkout
dsh plugin --profile web add ./dsh-tesseract-ocr-0.3.2.tgz
dsh plugin --profile web add github:maxwell-feng/dsh-tesseract-ocr

Git installs fetch sources, not built artifacts: the package's prepare script runs tsc to rebuild lib/ from source, and pnpm ≥ 10 requires you to allow the build once (it prints the exact pnpm-workspace.yaml snippet).

npm install registers the tesseract-ocr row by itself. The package ships a bundle patch (dsh.bundle + its own cordis.patch.yml) that inserts the tesseract-ocr loader entry. Do not also add a manual - insert: row with the same id to your profile — dsh 0.1.2-rc.1 rejects duplicate loader entry ids and dsh web fails to boot with duplicate loader entry id: tesseract-ocr.

Quick install via an AI agent

Hand this repository to any AI agent, or paste the instruction below, and the agent will install and verify the plugin for you:

Please install the dsh plugin in this repository by following https://github.com/maxwell-feng/dsh-tesseract-ocr/blob/main/agents-install.md. Run every preflight check, choose an install mode, then complete the mandatory verification: attach an image to a text-only model session and confirm the model answers with the recognized text.

agents-install.md is a step-by-step guide written for AI agents: preflight checks (including installing Tesseract and language packs), both install modes (permanent profile patch / temporary --patch overlay), mandatory functional verification, and troubleshooting for the failure modes you are likely to hit. Manual install instructions are below.

Why a plugin (not a skill)

dsh skills are Markdown instruction files injected into the model context — they cannot execute code, cannot hook the request pipeline, and cannot stop an image from being serialized. This feature needs exactly that, so it is a cordis plugin that hooks two public seams of the llm service (same design as windows-ocr):

1. Capability shim — ctx.llm.resolveModelInfo (also listModels). The host gates image attachments on inputModalities.includes("image") at three places: message admission, model switching, and the read_image tool. The shim answers "yes", so text models admit images.
2. Pre-step rewrite — agent/pre-step, the harness's official seam for replacing the messages that enter a model call ("Reject a proposed step or replace the messages that enter it"). Every image content block is replaced with an OCR text block before the request is built, so no attachment bytes are ever serialized and no image_url is ever built. It covers every dispatch path — ctx.llm.stream and prepareCall().stream both build from the step's messages; wrapping adapter.stream no longer works because the bundled adapters override prepareCall() and dispatch through generation-bound closures.

you attach an image
  → admission asks ctx.llm.resolveModelInfo (shimmed: "image" ✓)
  → image stored in the local attachment store (session log, UI preview)
  → agent loop proposes a step → agent/pre-step (rewritten)
  → image block read locally (ctx.attachments.readImage) → tesseract CLI
  → block replaced with <image_ocr>…text…</image_ocr>
  → request built from OCR'd messages → adapter serializes text only → provider

Requirements (Ubuntu)

sudo apt update
sudo apt install -y tesseract-ocr tesseract-ocr-chi-sim   # chi-sim = Simplified Chinese; add more packages as needed
tesseract --version        # verify
tesseract --list-langs     # verify installed languages

Language packs: tesseract-ocr-eng (usually pulled in by the base package), tesseract-ocr-chi-sim, tesseract-ocr-chi-tra, tesseract-ocr-jpn, … The language config joins multiple tags with +, e.g. eng+chi_sim.

Install into dsh

Installing via an AI agent

agents-install.md in this repository is a step-by-step installation guide written for AI agents (and careful humans). Give it to an agent — e.g. "install this plugin per agents-install.md from https://github.com/maxwell-feng/dsh-tesseract-ocr" — and the agent can perform the preflight checks, install, verification, and troubleshooting on its own. The guide covers both install modes, the mandatory functional verification (attach an image → model answers with the OCR text), and the failure modes you are likely to hit.

Manual install

Two official ways to load this plugin, both referencing the plugin file by absolute path (see docs/user/develop/basic). On Windows the path must be a file:// URL — a bare C:/... path is parsed as the c: URL scheme and the loader rejects it. On Linux a plain absolute path works too:

name: '/home/you/tesseract-ocr/lib/index.js'

Permanent: profile patch layer

Append to your profile's cordis.patch.yml (e.g. ~/.dsh/profiles/web/cordis.patch.yml):

- insert:
    - id: tesseract-ocr
      name: '/home/you/tesseract-ocr/lib/index.js'
      config:
        language: eng+chi_sim
        passthrough: false

Then restart dsh web. Remove the rows to uninstall — the plugin restores the original llm / adapter methods on unload.

Choose one way to load the plugin: the npm bundle (above) or this manual insert — never both. Both register the same tesseract-ocr entry id, and dsh 0.1.2-rc.1 fails the boot with duplicate loader entry id: tesseract-ocr when the row exists twice. If the row is already present (for example after an npm bundle install), configure it with an id-targeted override row instead of inserting a second one.

Temporary: --patch overlay

Put the same rows in an overlay file and boot with it; your profile stays untouched:

dsh --profile web --patch /home/you/tesseract-ocr/dev.patch.yml

Notes

• dsh web failing with EADDRINUSE means an older instance still holds the port: ss -ltnp | grep 3080, stop that process, start again.
• For a packaged install (npm / tarball / github:user/repo), package the plugin as a bundle (dsh.bundle + cordis.patch.yml, see docs/user/develop/basic/publish); a git install additionally needs a prepare build script and pnpm allowBuilds consent.

Configuration

All settings live in the patch row tesseract-ocr. Configuration is validated at load time (Schemastery Config schema) — an invalid value fails the boot with an actionable error instead of being silently ignored:

Usage

Attach any image to a text-model session and send a message — the plugin intercepts agent/pre-step, OCRs the image locally via the tesseract CLI, and replaces the image block with a text block before the request is built. No code or model-config changes needed; every provider/model in dsh benefits.

How the model sees the image

Each image block becomes a text block (local filenames are not forwarded):

<image_ocr>
…recognized lines…
</image_ocr>

Recognition text is cached per attachment id for the lifetime of the dsh process, so repeated turns do not re-run OCR.

Temp-file hygiene

Every OCR run writes its input image into a fresh temporary directory (tesseract-ocr-* under the system temp dir). On timeout the child process is terminated and awaited before cleanup. The directory is removed in finally (with one retry and a warning log on failure) — on success, on OCR error, and on timeout — so no per-run image file survives. At plugin start, any orphaned tesseract-ocr-* directories left behind by a previously crashed process are swept as well. Nothing is written outside the plugin's own temporary directory and the dsh attachment store.

Smoke test (no dsh needed)

# render a test image with text, then OCR it
convert -size 400x120 xc:white -pointsize 36 -fill black \
  -draw "text 20,80 'Hello OCR 123'" /tmp/ocr-test.png   # ImageMagick; any PNG works
tesseract /tmp/ocr-test.png stdout -l eng --psm 3

Exit 0 with the recognized text means Tesseract is ready.

Verification inside dsh

1. Attach an image to a text-model session and send a message — the model should answer using the recognized text.
2. Confirm the image never goes out: open DevTools → Network in the web UI, inspect the request to your provider base URL, and verify the payload contains only text content parts (no image_url / data URI).

Uninstall

dsh plugin --profile web remove dsh-tesseract-ocr

For manual installs, delete the tesseract-ocr row from your profile's cordis.patch.yml and restart dsh --profile web. The plugin restores the original llm shims on unload; a full restart is safest after removal. After uninstall, text-model image attachments are refused again (fail-closed).

Limitations

• Recognition quality depends on the installed language packs and psm; tune language/psm per use case.
• Image formats depend on the Tesseract/Leptonica build: PNG/JPEG/TIFF/BMP are safe; WebP/GIF may require additional Leptonica support.
• Cache is per process; a long-lived session keeps OCR text cached, bounded by maxCacheEntries.
• The plugin registers one fiber-scoped agent/pre-step listener and restores the llm capability shims on unload. A full restart is still the safest path after any dsh update.
• If the plugin is removed, image attachments to text models are refused again (fail-closed), not uploaded.
• Package name on npm is dsh-tesseract-ocr (unscoped) to avoid colliding with the unrelated tesseract-ocr package.

License

MIT