Obisidian

[cold vector]

I have a new openclaw setup help me setup obisidian memory for my new openclaw agent

[vernal robin]

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

[sturdy bone]

Yep, easiest setup is memory-wiki in Obsidian mode.

First, send me:

openclaw --version

Then add this to your openclaw.json:

{
  plugins: {
    entries: {
      "memory-wiki": {
        enabled: true,

        config: {
          vaultMode: "isolated",
          vault: {
            path: "~/Obsidian/OpenClaw-Wiki",
            renderMode: "obsidian",
          },
          obsidian: {
            enabled: true,
            useOfficialCli: false, // turn on later if you installed the official obsidian CLI
            vaultName: "OpenClaw Wiki",
            openAfterWrites: false,
          },
          ingest: {
            autoCompile: true,
            maxConcurrentJobs: 1,

            allowUrlIngest: true,
          },
          search: {
            backend: "local",
            corpus: "wiki",
          },
          render: {
            preserveHumanBlocks: true,
            createBacklinks: true,
            createDashboards: true,
          },
        },
      },
    },
  },

}

Then run:

openclaw gateway restart
openclaw wiki init
openclaw wiki status

Important bit: this does not replace normal OpenClaw memory. Your agent still uses MEMORY.md and daily notes, and memory-wiki gives you an Obsidian-friendly knowledge vault beside that.

If you already use plugins.allow, add "memory-wiki" there too.

Docs:

• https://docs.openclaw.ai/plugins/memory-wiki
• https://docs.openclaw.ai/cli/wiki
• https://docs.openclaw.ai/concepts/memory

If you paste your current openclaw.json, I can trim this to the exact config you need.

[cold vector]

claude cli token auth not working ⚠️ Agent failed before reply: All models failed (4): anthropic/claude-haiku-4-5: Provider anthropic has auth issue (skipping all models) (auth) | anthropic/claude-sonnet-4-6: Provider anthropic has auth issue (skipping all models) (auth) | anthropic/claude-opus-4-6: Provider anthropic has auth issue (skipping all models) (auth) | anthropic/claude-opus-4-7: Provider anthropic has auth issue (skipping all models) (auth).

[sturdy bone]

That usually means OpenClaw is trying to use Anthropic for this agent, but it cannot find a usable Anthropic credential on the gateway host.

Also important: new agents do not inherit the main agent’s Anthropic auth.

For a fresh setup, I would not use legacy Anthropic token auth if you can avoid it. Use one of these instead:

If you want to reuse Claude CLI login

Run on the gateway machine:

openclaw --version
claude --version
openclaw models auth login --provider anthropic --method cli --set-default
openclaw models status --probe

If you want direct Anthropic API auth

Put your key in ~/.openclaw/.env on the gateway machine:

ANTHROPIC_API_KEY=sk-ant-...

Then run:

openclaw models auth order clear --provider anthropic
openclaw gateway restart
openclaw models status --probe

A couple of gotchas that bite people a lot:

• Run this on the gateway host, not your laptop, if you are in remote mode.

• If the gateway runs via systemd/launchd, shell env vars often do not get inherited, so ~/.openclaw/.env is safer than exporting in your shell.
• If you pinned a bad Anthropic auth profile before, openclaw models auth order clear --provider anthropic helps.

Docs:

• https://docs.openclaw.ai/providers/anthropic
• https://docs.openclaw.ai/cli/models
• https://docs.openclaw.ai/help/faq

If you paste these two outputs, I can tell you the exact next fix:

openclaw --version
openclaw models status --probe

Before flagging it, I’d verify whether this is just per-agent Anthropic auth. New agents do not inherit the main agent’s creds.

From the gateway host, please post:

openclaw --version
openclaw models status --probe

And if they want Claude CLI reuse:

openclaw models auth login --provider anthropic --method cli --set-default

If that still fails after re-auth, the right escalation path is a GitHub issue, not a Discord invite: