LumeAPIShort path: Cheap LLM API · LLM API gateway · Multi-model API · Models · AI API pricing · Models
Last verified: July 26, 2026
Register a custom provider under models.providers, set its baseUrl to https://api.lumeapi.site/v1, choose the openai-completions API adapter, and list each allowed LumeAPI model id explicitly. Then set the primary agent model to a provider/model reference such as lumeapi/gpt-5.6-terra. OpenClaw’s current documentation distinguishes generic OpenAI-compatible proxy routes from native OpenAI/Codex routes, so use a custom provider id rather than pretending LumeAPI is the built-in OpenAI service.
LumeAPI is an independent third-party gateway and is not affiliated with OpenClaw or any model provider.
Quick answer
| Question | Answer |
|---|---|
| Is OpenClaw just a coding IDE? | No; it is a personal-assistant runtime with gateway, channels, agents, tools, and providers |
| Where is a custom provider registered? | models.providers.<provider-id> |
| Which API adapter fits LumeAPI’s published catalog? | openai-completions |
| What base URL should be tested? | https://api.lumeapi.site/v1 |
| What model ref does the agent select? | <provider-id>/<catalog-id>, such as lumeapi/gpt-5.6-terra |
In short
Register a custom provider under models.providers, set its baseUrl to https://api.lumeapi.site/v1, choose the openai-completions API adapter, and list each allowed LumeAPI model id explicitly. Then set the primary agent model to a provider/model reference such as lumeapi/gpt-5.6-terra.
OpenClaw architecture in one picture
Telegram / web / other channel
│
▼
OpenClaw gateway
│
agent + tools + memory
│
model provider policy
│
▼
LumeAPI OpenAI-compatible API
├─ gpt-5.4-mini
├─ gpt-5.6-terra
├─ gpt-5.6-sol
├─ claude-sonnet-4-6
└─ claude-opus-4-8Unlike a one-off coding assistant, an OpenClaw installation may remain available across channels and recurring tasks. Cost therefore depends on daily message volume, tool loops, memory retrieval, retries, and background behavior.
Candidate custom-provider configuration
OpenClaw’s current provider documentation requires a provider definition plus explicit model entries. A staging configuration can look like this:
{
env: {
LUMEAPI_API_KEY: "sk-your-lumeapi-key"
},
models: {
providers: {
lumeapi: {
baseUrl: "https://api.lumeapi.site/v1",
apiKey: "${LUMEAPI_API_KEY}",
api: "openai-completions",
models: [
{ id: "gpt-5.4-mini", name: "GPT-5.4 mini" },
{ id: "gpt-5.6-terra", name: "GPT-5.6 Terra" },
{ id: "gpt-5.6-sol", name: "GPT-5.6 Sol" },
{ id: "claude-sonnet-4-6", name: "Claude Sonnet 4.6" }
]
}
}
},
agents: {
defaults: {
model: { primary: "lumeapi/gpt-5.6-terra" }
}
}
}Do not commit a real key. Confirm the installed OpenClaw version accepts environment interpolation in this exact location; otherwise use its documented secret-reference mechanism.
Why the custom provider id matters
OpenClaw treats native OpenAI, Codex OAuth, Azure, and generic compatible proxies differently. Native routes may receive OpenAI-only request shaping, transport behavior, service tiers, or Codex runtime selection. A LumeAPI route is a compatible proxy route and should use the compatible adapter and looser proxy behavior.
The provider prefix is local routing metadata. OpenClaw strips it before sending the configured model id to the upstream provider, so lumeapi/gpt-5.6-terra should produce upstream model id gpt-5.6-terra when the configuration is correct.
Route GPT and Claude by assistant task
| Task | First route | Escalation trigger |
|---|---|---|
| Classification, short summaries, routine channel replies | gpt-5.4-mini | Repeated format or accuracy failure |
| General assistant and lightweight coding | gpt-5.6-terra | Complex multi-tool task |
| Difficult reasoning or coding workflow | gpt-5.6-sol | Default only when acceptance value supports it |
| Claude-based long-form or coding experiment | claude-sonnet-4-6 | Measured task-specific failure |
| Premium Claude fallback | claude-opus-4-8 | Explicitly approved high-value task |
OpenClaw model selection should be governed by an allowlist. Do not let untrusted channel content choose arbitrary provider/model references.
What does an always-on assistant cost?
Assume 600,000 input and 80,000 output tokens per day across all channels and tools:
| Route | Calculated daily cost | 30-day cost |
|---|---|---|
| GPT-5.4 mini | $0.243 | $7.29 |
| GPT-5.6 Terra | $0.810 | $24.30 |
| GPT-5.6 Sol | $1.620 | $48.60 |
| Claude Sonnet 4.6 | $1.500 | $45.00 |
| Claude Opus 4.8 | $2.500 | $75.00 |
These are token scenarios, not observed OpenClaw bills. Memory embeddings, media, retries, background jobs, gateway overhead, and other services may add cost. The real KPI is cost per completed assistant task and the percentage of spend caused by retries or unattended loops.
What most setup guides get wrong
OpenClaw is not a drop-in replacement for Claude Code. It is a broader runtime that connects channels, tools, memory, agents, and model providers. Its always-on nature changes both security and cost.
A second mistake is using the built-in openai route with a custom base URL and assuming every native Codex feature remains available. OpenClaw explicitly separates native and proxy-compatible behavior. A third mistake is registering a model alias without adding the corresponding model under models.providers; an alias alone does not create a runtime route.
Production controls
- Store keys using OpenClaw’s secret system.
- Restrict allowed models and tools by agent/channel.
- Cap turns and timeouts for unattended tasks.
- Log provider, upstream model id, tokens, retries, and final task status.
- Test code blocks and backticks through any WAF in front of the compatible endpoint.
- Fail closed if a model id disappears or a provider schema changes.
Use quality/cost routing guidance and the multi-model production checklist to define the policy.
What most guides get wrong
The usual mistake on How Do You Configure OpenClaw to Use LumeAPI as a Multi-Model Backend? is to chase the lowest posted per-million-token rate while ignoring orchestration: repeated context, oversized tool output, silent retries, and premium-model defaults often dominate the invoice.
A cheaper list price does not help when every agent hop resends the same repository map or when failed tests trigger unbounded re-runs. Optimize cost per successful task—not the price of a single completion—before you renegotiate vendor contracts or swap gateways.
A realistic production scenario
A six-person platform team standardized on one premium model for How Do You Configure OpenClaw to Use LumeAPI as a Multi-Model Backend? after a strong demo score. Early invoices looked fine; by week three, failing test loops had doubled input tokens without more accepted output.
They kept the orchestrator but tightened routing: a lower-cost catalog model handled exploration, a mid-tier model handled implementation, and premium routes required a logged failure condition. They truncated test logs and stopped attaching full diffs on every retry.
The breakthrough was reconciling LumeAPI usage records with task ids. Token list price mattered less than eliminating duplicate context across retries.
Expert take
LumeAPI is an independent third-party OpenAI-compatible gateway—not OpenAI, Anthropic, or Google. Published catalog rates can be materially below official reference pricing, but compatibility does not guarantee identical behavior for every native feature, streaming edge case, tool schema, Batch workflow, or prompt-cache semantics.
Follow this advice when you control prompts, retries, and routing and need auditable USD billing across GPT, Claude, and Gemini on one key. Do not cut over blindly if you depend on provider-native capabilities your stack has not tested end to end.
Run a bounded smoke test on your exact SDK, model id, streaming mode, and tool path; reconcile usage logs with orchestrator task ids; and keep a fallback route until error rates and latency are acceptable.
FAQ
Does this guide apply to every LumeAPI model?
Verify the exact model id and rate on /models before production use for How Do You Configure OpenClaw to Use LumeAPI as a Multi-Model Backend?.
Can OpenClaw use an OpenAI-compatible proxy?
Yes. Its current docs describe custom providers with base URLs and an openai-completions adapter.
Can one provider expose GPT and Claude ids?
Yes if the upstream gateway exposes both through the same compatible contract. Each model must be registered and tested.
Why not use openai/gpt-5.6-sol?
That prefix belongs to OpenClaw’s built-in provider behavior. Use lumeapi/gpt-5.6-sol for the custom LumeAPI provider defined in this article.
Sources and testing boundary
- OpenClaw model provider documentation
- OpenClaw repository — 384,150 stars observed July 26, 2026
- LumeAPI model catalog
No billable OpenClaw-to-LumeAPI run was performed for this report. Before production use, verify openclaw models list, a channel message, one tool call, the upstream model id, and the matching LumeAPI usage record.