Paperclip + Z.AI GLM Coding Plan

The missing setup guide for running Paperclip AI agents on Z.AI's GLM Coding Plan subscription.

If you just bought a Z.AI GLM Coding Plan and your Paperclip agents are throwing error 1113: Insufficient balance or no resource package — or the model validator is rejecting everything except github-copilot/* — this is the fix. Took me most of a day to figure out. Hope it saves you the time.

Disclosure: the Z.AI link above is my affiliate link. Same $18/month for you either way; I get a small kickback if you use it. If you'd rather skip it, just go to z.ai directly — the fix below works either way.

---

TL;DR — three things you'll probably get wrong

Wrong	Right
Endpoint: api.z.ai/api/paas/v4	Endpoint: api.z.ai/api/coding/paas/v4
Env var: OPENAI_API_KEY / ZAI_API_KEY	Env var: ZHIPU_API_KEY
Model: glm-5.1	Model: zai-coding-plan/glm-5.1

Fix all three and it just works.

The working config

In Paperclip's agent configuration (either via the UI or the agents.adapter_config JSON):

{
  "model": "zai-coding-plan/glm-5.1",
  "env": {
    "ZHIPU_API_KEY": {
      "type": "plain",
      "value": "<your-z.ai-api-key>"
    }
  }
}

No OPENAI_BASE_URL override. No source patches. No middleware. Paperclip's OpenCode adapter already knows about the zai-coding-plan provider — you just need to give it the right env var.

Why each of those three things matters

Endpoint: two separate billing paths

Z.AI runs two API endpoints that look nearly identical but bill differently:

• https://api.z.ai/api/paas/v4 — pay-as-you-go, needs pre-loaded credits
• https://api.z.ai/api/coding/paas/v4 — the Coding Plan subscription

Even with an active subscription, your balance at the pay-as-you-go endpoint is $0. Hit it and you get error 1113 every time, even though your dashboard says the plan is active.

Env var: OpenCode has its own provider catalog

I spent hours trying to trick OpenCode into treating Z.AI as an OpenAI-compatible endpoint by setting OPENAI_BASE_URL=https://api.z.ai/.... OpenCode doesn't work that way. It ships with a baked-in providers catalog, and each provider expects its own env var. The Z.AI provider expects ZHIPU_API_KEY. Setting OPENAI_API_KEY does nothing for Z.AI calls.

Model prefix: OpenCode has three Z.AI providers

OpenCode's provider catalog actually contains three separate Z.AI entries, each pointing at a different endpoint:

zai/*               →  api.z.ai/api/paas/v4         (pay-as-you-go)
zai-coding-plan/*   →  api.z.ai/api/coding/paas/v4  (subscription) ← this one
zhipuai/*           →  open.bigmodel.cn/api/paas/v4 (China-mainland)

Your agent's model field needs the zai-coding-plan/ prefix. Passing just glm-5.1 fails because OpenCode has no idea which provider to route to.

Verify it works before pointing real agents at it

Shell into your Paperclip server container and run:

ZHIPU_API_KEY=<your-key> opencode run \
  --model zai-coding-plan/glm-5.1 \
  --format json \
  "Reply with just the word hello."

Expected: a text event containing "hello", "cost": 0, and (nice bonus) "cache.read" showing thousands of tokens. That's confirmation that:

1. Your auth works
2. The subscription is active and billing correctly
3. Z.AI's native cache layer is engaged

If you still see error 1113, your subscription hasn't propagated yet. Community reports suggest 5–15 minutes after purchase, even though Z.AI's dashboard marks the plan as active immediately. Wait a bit and retry.

Bulk-switching all your agents at once

Paperclip's UI only lets you change models one agent at a time. If you're running a dozen+ agents and want to flip them all at once — or toggle back to OpenRouter for failover — there's a helper script at scripts/paperclip-switch-provider.sh.

Set your API keys at the top of the file, then:

./paperclip-switch-provider.sh zai              # Switch everything to zai-coding-plan/glm-5.1
./paperclip-switch-provider.sh zai glm-5        # Specific model
./paperclip-switch-provider.sh openrouter       # Fail back to openrouter/z-ai/glm-5.1
./paperclip-switch-provider.sh --status         # See what every agent is currently on

It updates the adapter_config JSONB column directly in your Paperclip Postgres database, handling env var cleanup between providers so you never end up with stale keys confusing OpenCode.

What --status looks like once it's working

Current provider/model for each agent:

               agent                |          model          |     provider
------------------------------------+-------------------------+-------------------
 ALE/Affiliate Marketing Agent      | zai-coding-plan/glm-5.1 | zai (coding plan)
 ALE/CEO                            | zai-coding-plan/glm-5.1 | zai (coding plan)
 ALE/Community Retention Manager    | zai-coding-plan/glm-5.1 | zai (coding plan)
 ALE/Content Ops Editor             | zai-coding-plan/glm-5.1 | zai (coding plan)
 ALE/Founding Engineer              | zai-coding-plan/glm-5.1 | zai (coding plan)
 ALE/Funnel Automation Engineer     | zai-coding-plan/glm-5.1 | zai (coding plan)
 ALE/Growth Offer Strategist        | zai-coding-plan/glm-5.1 | zai (coding plan)
 ALE/OpenRouter Model Auditor       | zai-coding-plan/glm-5.1 | zai (coding plan)
 ALE/Partnerships Affiliate Manager | zai-coding-plan/glm-5.1 | zai (coding plan)
 ALE/Substack Publisher             | zai-coding-plan/glm-5.1 | zai (coding plan)
 MIR/CEO                            | zai-coding-plan/glm-5.1 | zai (coding plan)
 MIR/Founding Engineer              | zai-coding-plan/glm-5.1 | zai (coding plan)
 MIR/Growth Lead                    | zai-coding-plan/glm-5.1 | zai (coding plan)
 PIC/COO                            | zai-coding-plan/glm-5.1 | zai (coding plan)
 PIC/Content & SEO Strategist       | zai-coding-plan/glm-5.1 | zai (coding plan)
 PIC/Founding Engineer              | zai-coding-plan/glm-5.1 | zai (coding plan)
 PIC/Sales & Growth Lead            | zai-coding-plan/glm-5.1 | zai (coding plan)
 PIC/Support & Knowledge Manager    | zai-coding-plan/glm-5.1 | zai (coding plan)

(Real output from my install running 3 companies and ~20 agents, all happily billing against one flat subscription.)

Why it's worth the setup hassle

Coding Plan is a flat-rate subscription. Once the wiring works, every API call is effectively free at the margin. You can point the full firehose of Paperclip's heartbeat runs at zai-coding-plan/glm-5.1 without watching a per-token meter.

Z.AI's native cache layer is included — I see 7-8K tokens cache-hit on repeat runs out of the box, which cuts effective latency and makes the subscription even more of a bargain.

Compared to openrouter/z-ai/glm-5.1 (same underlying model, routed through OpenRouter's per-token markup), the subscription is clearly the right call for anyone doing real heartbeat-driven agent work. I went from ~$5/day in OpenRouter token spend back down to effectively $0.

Related projects

• Paperclip — the AI agent orchestration platform this fix is for
• paperclip-vision by Aron Prins — a Claude Code skill that interviews you and produces VISION.md + CEO_BOOTSTRAP.md so your Paperclip CEO agent has a real mandate instead of guessing. Nice complement to this one — fix your model routing here, fix your agent strategy there. Aron's an active Paperclip contributor and worth following.
• Z.AI GLM Coding Plan — $18/month flat rate for GLM-5.1 and siblings (affiliate link — same price for you, small kickback for me)
• OpenCode — the CLI adapter Paperclip uses; the zai-coding-plan provider definition lives in its models catalog

Who I am

I'm Nicholas. I write about building and running AI-assisted businesses at nicholasrhodes.substack.com and I'm building MirrorMemory. If this doc saved you an afternoon, come say hi.

License

MIT. Use it however you want. Pull requests welcome if Z.AI changes anything.