Can i Connect sambanova free models instead of google api | Friends of the Crustacean 🦞🤝 | Page 1

manic thunder Apr 19, 2026, 8:01 PM

#

Can i Connect sambanova free models instead of google api?

hushed iceBOT Apr 19, 2026, 8:01 PM

#

celest swiftBOT Apr 19, 2026, 8:01 PM

#

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

barren spindleBOT Apr 19, 2026, 8:04 PM

#

Yep, for normal chat models, yes.

I don’t see a built-in SambaNova provider in the docs, but OpenClaw supports custom OpenAI-compatible endpoints, so if SambaNova gives you an API key + endpoint, you can use it instead of Google.

Fast path:

openclaw onboard --non-interactive \
  --auth-choice custom-api-key \
  --custom-base-url "<your SambaNova API URL>" \
  --custom-model-id "<model-id>" \
  --custom-provider-id "sambanova" \
  --custom-compatibility openai

Or add it manually under models.providers.

#

Docs:
https://docs.openclaw.ai/cli/onboard
https://docs.openclaw.ai/gateway/configuration-reference#modelsproviders

One catch, this only works if SambaNova’s “free” offering includes actual API access. If it’s just a web playground, then no.

If you want, paste openclaw --version and the SambaNova model/endpoint you’re targeting, and I’ll turn it into an exact config snippet.

manic thunder Apr 19, 2026, 8:56 PM

#

"⚠️ Context limit exceeded. I've reset our conversation to start fresh - please try again.

To prevent this, increase your compaction buffer by setting agents.defaults.compaction.reserveTokensFloor to 20000 or higher in your config."
Im using github models gpt-4.1, ive added ````agents.defaults.compaction.reserveTokensFloor``` and still same issue

barren spindleBOT Apr 19, 2026, 8:58 PM

#

That warning is generic, so if you already set agents.defaults.compaction.reserveTokensFloor, it does not necessarily mean that setting is the real problem.

A few quick checks:

What does openclaw --version show?
Did you restart the gateway after changing the config?
Are you using the built-in github-copilot/gpt-4.1 path, or a custom GitHub Models/OpenAI-compatible provider?

Why this matters:

reserveTokensFloor only changes the buffer before compaction. It does not increase the model’s actual context window.
If the provider/model has a missing or wrong contextWindow, compaction can still fail.
Older builds had compaction edge cases, so version matters a lot here.

Also worth trying if you haven’t already:

#

{
  agents: {
    defaults: {
      compaction: {
        mode: "safeguard",
        reserveTokensFloor: 20000
      }
    }
  }
}

Docs:
https://docs.openclaw.ai/concepts/compaction
https://docs.openclaw.ai/gateway/configuration-reference#agentsdefaultscompaction

#

If you paste:

openclaw --version
your agents.defaults.compaction block
and whether this is github-copilot or a custom GitHub Models provider

…I can narrow it down pretty fast.

manic thunder Apr 19, 2026, 9:06 PM

#

look my config:

  "agents": {
    "defaults": {
       "compaction": { "reserveTokensFloor": 30000 },
      "workspace": "/root/.openclaw/workspace",
      "model": {
        "primary": "github/gpt-4.1"
      },
      "models": {
        "github/openai/gpt-5-mini": {},
        "github/openai/gpt-4.1": {}
      }
    }
  },```


``` {
            "id": "gpt-4.1",
            "name": "gpt-4.1 (Custom Provider)",
            "contextWindow": 128000,
            "maxTokens": 4096,
            "input": [
              "text"
            ],
            "cost": {
              "input": 0,
              "output": 0,
              "cacheRead": 0,
              "cacheWrite": 0
            },
            "reasoning": false
          }```

barren spindleBOT Apr 19, 2026, 9:07 PM

#

I think I see one config bug right away.

Your refs are inconsistent:

model.primary = github/gpt-4.1
but agents.defaults.models uses github/openai/gpt-4.1

OpenClaw model refs are provider/model, so those keys should match the actual provider id and model id exactly.

Try this shape instead:

{
  agents: {
    defaults: {
      workspace: "/root/.openclaw/workspace",

#

      model: {
        primary: "github/gpt-4.1"
      },
      compaction: {
        mode: "safeguard",
        reserveTokensFloor: 30000
      },
      models: {
        "github/gpt-4.1": {},
        "github/gpt-5-mini": {}
      }
    }
  }
}

#

A couple important notes:

reserveTokensFloor is only a buffer, it does not increase the real model limit.
If the upstream GitHub model limit is lower than what your custom provider says, 30000 can actually make compaction happen earlier.
After changing config, restart the gateway.

If it still breaks after fixing the model refs + restart, send:

openclaw --version
your models.providers.github block

Then we can check whether the custom provider metadata, especially contextWindow, matches what GitHub Models actually accepts.

Docs:
https://docs.openclaw.ai/gateway/configuration-reference#agentsdefaultscompaction
https://docs.openclaw.ai/gateway/configuration-reference#modelsproviders

manic thunder Apr 19, 2026, 9:16 PM

#

  "agents": {
    "defaults": {
       "compaction": {
           "mode": "safeguard",
          "reserveTokensFloor": 30000
},
      "workspace": "/root/.openclaw/workspace",
      "model": {
        "primary": "github/gpt-4.1"
      },
      "models": {
        "github/gpt-5-mini": {},
        "github/gpt-4.1": {}
      }
    }
  },```

```"github": {
        "baseUrl": "https://models.inference.ai.azure.com",
        "api": "openai-completions",
        "apiKey": "...",
        "models": [
          {
            "id": "openai/gpt-5-mini",
            "name": "openai/gpt-5-mini (Custom Provider)",
            "contextWindow": 16000,
            "maxTokens": 4096,
            "input": [
              "text"
            ],
            "cost": {
              "input": 0,
              "output": 0,
              "cacheRead": 0,
              "cacheWrite": 0
            },
            "reasoning": false
          },
           {
            "id": "gpt-4.1",
            "name": "gpt-4.1 (Custom Provider)",
            "contextWindow": 128000,
            "maxTokens": 4096,
            "input": [
              "text"
            ],
            "cost": {
              "input": 0,
              "output": 0,
              "cacheRead": 0,
              "cacheWrite": 0
            },
            "reasoning": false
          }
        ]```

barren spindleBOT Apr 19, 2026, 9:17 PM

#

I need to correct one thing from my last reply, I oversimplified the model refs.

With a custom provider, the model id is whatever the upstream expects. For GitHub Models, refs like openai/gpt-4.1 are normal, so github/openai/gpt-4.1 can be valid.

What I see now is inconsistent model IDs:

gpt-5-mini is stored as openai/gpt-5-mini
gpt-4.1 is stored as just gpt-4.1

I’d make them consistent and match GitHub Models exactly, like this:

{
  agents: {
    defaults: {
      workspace: "/root/.openclaw/workspace",

#

      compaction: {
        mode: "safeguard",
        reserveTokensFloor: 30000
      },
      model: {
        primary: "github/openai/gpt-4.1"
      },
      models: {
        "github/openai/gpt-4.1": {},
        "github/openai/gpt-5-mini": {}
      }
    }
  },
  models: {
    providers: {

#

      github: {
        baseUrl: "https://models.inference.ai.azure.com",
        api: "openai-completions",
        apiKey: "...",
        models: [
          {
            id: "openai/gpt-4.1",
            name: "openai/gpt-4.1 (Custom Provider)",
            contextWindow: 128000,
            maxTokens: 4096,
            input: ["text"],
            cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
            reasoning: false
          },
          {

#

            id: "openai/gpt-5-mini",
            name: "openai/gpt-5-mini (Custom Provider)",
            contextWindow: 16000,
            maxTokens: 4096,
            input: ["text"],
            cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
            reasoning: false
          }
        ]
      }
    }
  }
}

Then restart the gateway.

#

I’d fix that first before changing anything else. If it still throws the same compaction error after that, send:

openclaw --version
the exact error from gateway logs around the overflow

Docs:
https://docs.openclaw.ai/gateway/configuration-reference#modelsproviders
https://docs.openclaw.ai/gateway/configuration-reference#agentsdefaultscompaction

Glad you pasted the provider block, that made the mismatch a lot clearer.

manic thunder Apr 19, 2026, 9:23 PM

#

400 Unknown model: openai/gpt-4.1 openclaw is texting

barren spindleBOT Apr 19, 2026, 9:23 PM

#

Yep, that confirms it. This endpoint wants bare model IDs, not openai/....

So for your GitHub provider, make them all consistent like this:

{
  agents: {
    defaults: {
      workspace: "/root/.openclaw/workspace",
      compaction: {
        mode: "safeguard",
        reserveTokensFloor: 30000
      },
      model: {
        primary: "github/gpt-4.1"
      },

#

      models: {
        "github/gpt-4.1": {},
        "github/gpt-5-mini": {}
      }
    }
  },
  models: {
    providers: {
      github: {
        baseUrl: "https://models.inference.ai.azure.com",
        api: "openai-completions",
        apiKey: "...",
        models: [
          {
            id: "gpt-4.1",

#

            name: "gpt-4.1 (Custom Provider)",
            contextWindow: 128000,
            maxTokens: 4096,
            input: ["text"],
            cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
            reasoning: false
          },
          {
            id: "gpt-5-mini",
            name: "gpt-5-mini (Custom Provider)",
            contextWindow: 16000,
            maxTokens: 4096,
            input: ["text"],
            cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
            reasoning: false

#

So the important fix is:

gpt-4.1, not openai/gpt-4.1
gpt-5-mini, not openai/gpt-5-mini

Then restart the gateway.

If you still get the context limit exceeded error after that, send openclaw --version, because then we’re probably looking at a compaction/version issue, not a model-name issue.

#

Docs:
https://docs.openclaw.ai/gateway/configuration-reference#modelsproviders
https://docs.openclaw.ai/gateway/configuration-reference#agentsdefaultscompaction

manic thunder Apr 19, 2026, 9:24 PM

#

not "primary": "github/openai/gpt-4.1" but "primary": "github/gpt-4.1"?

barren spindleBOT Apr 19, 2026, 9:24 PM

#

Yes, exactly, use:

"primary": "github/gpt-4.1"

Not:

"primary": "github/openai/gpt-4.1"

Because your endpoint just told us it accepts bare gpt-4.1 and rejects openai/gpt-4.1.

So keep it consistent everywhere:

primary: github/gpt-4.1

#

allowlist: github/gpt-4.1, github/gpt-5-mini
provider model ids: gpt-4.1, gpt-5-mini

Sorry for the back-and-forth, that 400 error was the missing clue.

manic thunder Apr 19, 2026, 9:26 PM

#

so "id": "gpt-5-mini", "name": "gpt-5-mini (Custom Provider)",

barren spindleBOT Apr 19, 2026, 9:26 PM

#

Yep 👍

{
  "id": "gpt-5-mini",
  "name": "gpt-5-mini (Custom Provider)"
}

And then match it everywhere else as:

github/gpt-5-mini in agents.defaults.models
github/gpt-4.1 as primary
gpt-5-mini / gpt-4.1 as the provider model ids

One small note, your gpt-5-mini entry is only 16000 context, so if you switch to that model you’ll hit compaction much sooner than gpt-4.1.

#

Glad we caught the naming mismatch. If this fixed it, please right-click the solution message → Apps → ✅ Mark Solution so Answer Overflow can index it.

manic thunder Apr 19, 2026, 9:29 PM

#

Context limit exceeded. I've reset our conversation to start fresh - please try again.

To prevent this, increase your compaction buffer by setting agents.defaults.compaction.reserveTokensFloor to 20000 or higher in your config.

barren spindleBOT Apr 19, 2026, 9:30 PM

#

Yeah, at this point that auto-reply is being misleading.

You already have reserveTokensFloor: 30000, so do not raise it further. The docs explicitly note that too-high compaction reserves can make compaction happen earlier.

Next thing I’d try is actually lowering the compaction budget for this custom provider, for example:

{
  agents: {
    defaults: {
      compaction: {
        mode: "safeguard",
        reserveTokensFloor: 0,
        reserveTokens: 12000,
        keepRecentTokens: 12000
      }

#

    }
  }
}

Then restart the gateway and test again.

Why:

reserveTokensFloor: 0 disables the enforced floor
lower reserveTokens / keepRecentTokens gives compaction more room to recover
with custom providers, the generic “set it to 20000+” message is not always correct

I also need your exact version now:

#Can i Connect sambanova free models instead of google api