#How can OpenClaw autonomously send images

1 messages · Page 1 of 1 (latest)

fierce geyser
#

How can OpenClaw autonomously send images after I provide it with the corresponding image paths and descriptions? TThese images serve as background knowledge and are automatically retrieved and sent during conversations.

eternal depotBOT
#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> âś… Mark Solution

fierce geyser
#

How can OpenClaw autonomously send images

tall moatBOT
#

If you mean “I tell the agent /path/to/img.png + a Discord channel:<id> and it posts it there”, there are two supported patterns:

1) Send media as part of an agent reply (same chat/thread)

Have the agent include a MEDIA:<path-or-url> line on its own line in its output:

Here you go.
MEDIA:./out/result.png

Notes:

2) Proactively send to a specific Discord channel id (any channel)

Use the outbound messaging pipeline (good for “send to channel X”):

#

CLI (easy to sanity-check first):

openclaw message send --channel discord \
  --target channel:1481950909914288158 \
  --media ./out/result.png \
  --message "caption (optional)"

Target format is channel:<id> on Discord. Docs: https://docs.openclaw.ai/cli/message

Common gotcha

The image path must exist on the Gateway host filesystem (or inside its sandbox/workspace). If your Gateway is on a VPS but the image path is on your laptop, OpenClaw can’t read it unless you upload it (URL) or move/copy it into the workspace on the gateway host.

If you tell me: (a) where the images live (local machine vs VPS), and (b) whether you want to post in this thread vs a different channel, I can point you at the cleanest exact setup.

#

“Autonomously send images during conversations” boils down to: store a searchable index (text → image path), then teach the agent (via a skill / AGENTS.md) to consult that index and attach matching images.

Two important constraints up front:

  1. OpenClaw does not “ingest” local images as knowledge automatically.
    If you want the content of the image to help answers, you need a text description/OCR summary stored somewhere (index file / memory). The model can’t use a random file path as knowledge by itself.

  2. To actually attach an image, the agent must output MEDIA:<path-or-url> (same chat) or use the outbound message tool (other chat).
    Docs: https://docs.openclaw.ai/start/openclaw#media-in-and-out

Recommended pattern: an “Image KB” index + skill

1) Put images somewhere the gateway can read

Put them inside the agent workspace (best) so paths are stable and allowed, e.g.:

#
  • ./kb/images/office-map.png
  • ./kb/images/product-arch.png

(If your gateway runs on a VPS, these files must exist on the VPS filesystem.)

2) Create an index file with descriptions/tags

Example ./kb/images/index.md:

# Image KB

- id: office-map
  description: Floor plan of the Vienna office with room names.
  tags: office, map, floorplan, vienna
  path: ./kb/images/office-map.png
# 
- id: product-arch
  description: High-level architecture diagram of the ingestion pipeline.
  tags: architecture, diagram, ingestion
  path: ./kb/images/product-arch.png

3) Add a skill (or AGENTS.md rule) that retrieves + attaches

Create a skill ./skills/image-kb/SKILL.md that instructs:

  • read ./kb/images/index.md
  • pick up to N relevant entries based on the user’s question
  • reply with a short caption + MEDIA:./kb/images/... lines

This makes the behavior consistent instead of “sometimes it remembers”.

Skill docs: https://docs.openclaw.ai/tools/skills

#

How the sending works in practice

Attach to the current conversation/thread

In the agent’s reply:

Here’s the architecture diagram for reference.
MEDIA:./kb/images/product-arch.png

OpenClaw will strip the MEDIA: line from visible text and send the image as an attachment.

Send to a different Discord channel

Then you need the outbound message tool / CLI (openclaw message send ...). Docs: https://docs.openclaw.ai/cli/message
(Also note cross-context restrictions; see FAQ: https://docs.openclaw.ai/help/faq#how-do-i-send-a-discord-message-from-telegram-crosscontext-messaging-denied)