feat: support ANTHROPIC_BASE_URL, ANTHROPIC_AUTH_TOKEN and SPACEBOT_MODEL env vars

[adryserage]

Summary

• Add ANTHROPIC_BASE_URL env var to override the Anthropic API endpoint
• Add ANTHROPIC_AUTH_TOKEN env var as an alternative to ANTHROPIC_API_KEY for proxy auth
• Add SPACEBOT_MODEL env var to override all process types with a single model
• Fix bug where build_anthropic_request() hardcoded the API URL, ignoring configured base_url

Why

Many organizations run Anthropic-compatible API proxies (LiteLLM, Azure AI Gateway, corporate proxies). This makes Spacebot usable in enterprise and self-hosted environments where direct API access isn't available.

New Environment Variables

Variable	Purpose	Default
ANTHROPIC_BASE_URL	Custom API endpoint URL	https://api.anthropic.com
ANTHROPIC_AUTH_TOKEN	Bearer token for proxy auth (fallback if no ANTHROPIC_API_KEY)	—
SPACEBOT_MODEL	Override all process types at once	—

These are non-breaking — existing ANTHROPIC_API_KEY behavior stays the same.

Changes

src/llm/anthropic/params.rs

• Remove hardcoded ANTHROPIC_API_URL constant
• Add base_url parameter to build_anthropic_request()
• Add messages_url() helper that appends /v1/messages to the base URL

src/llm/model.rs

• Pass provider_config.base_url to build_anthropic_request()

src/config.rs

• ANTHROPIC_AUTH_TOKEN as fallback for ANTHROPIC_API_KEY in both load_from_env() and from_toml()
• ANTHROPIC_BASE_URL overrides the default base URL when registering the Anthropic provider
• SPACEBOT_MODEL sets all routing process types (channel, branch, worker, compactor, cortex); specific vars like SPACEBOT_CHANNEL_MODEL take precedence

Bug Fix

Previously, build_anthropic_request() always posted to https://api.anthropic.com/v1/messages regardless of what base_url was configured in ProviderConfig. This meant custom base URLs set via [llm.provider.anthropic] TOML config were silently ignored. Now the provider's base_url is passed through and used.

Test plan

• Verify existing ANTHROPIC_API_KEY still works (no regression)
• Test ANTHROPIC_AUTH_TOKEN as sole credential
• Test ANTHROPIC_BASE_URL with a proxy endpoint
• Test SPACEBOT_MODEL overrides all process types
• Test SPACEBOT_CHANNEL_MODEL takes precedence over SPACEBOT_MODEL
• All 26 existing Anthropic tests pass

Closes #132

[pjv]

any chance you could also add:

• ANTHROPIC_MODEL
• CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC

[worldofgeese]

@adryserage @jamiepine running latest release. With the following config I'm receiving auth errors:

spacebot | 2026-02-23T07:22:45.804280Z ERROR handle_message: spacebot::agent::channel: message=channel LLM call failedchannel_id=discord:1464919871170023568:1475064317324824699error=CompletionError: ProviderError: Anthropic API error (403 Forbidden): unknown error channel_id=discord:1464919871170023568:1475064317324824699agent_id=mainmessage_id=1475392417254342849

[api]
bind = "::"

[llm]

[llm.providers.anthropic]
api_type = "anthropic"
base_url = "env:ANTHROPIC_BASE_URL"
api_key = "env:ANTHROPIC_AUTH_TOKEN"

[messaging.discord]
enabled = true
token = "env:DISCORD_BOT_TOKEN"

[[agents]]
id = "main"
default = true

[agents.routing]
channel = "anthropic/claude-opus-4-6"
branch = "anthropic/claude-opus-4-6"
worker = "anthropic/claude-sonnet-4-5"
compactor = "anthropic/claude-sonnet-4-5"
cortex = "anthropic/claude-opus-4-6"
voice = ""
rate_limit_cooldown_secs = 60

[defaults]
brave_search_key = "env:BRAVE_SEARCH_API_KEY"

[[bindings]]
agent_id = "main"
channel = "discord"
require_mention = true

This is also the case with variations such as:

[llm]
anthropic_key = "env:ANTHROPIC_AUTH_TOKEN"

[llm.providers.anthropic]
api_type = "anthropic"
base_url = "env:ANTHROPIC_BASE_URL"
api_key = "env:ANTHROPIC_AUTH_TOKEN"

or the [llm.providers.anthropic] block omitted entirely.

[worldofgeese]

I've been testing this PR with a corporate Anthropic-compatible proxy and I'm consistently getting 403 errors.

I believe the issue is that when ANTHROPIC_AUTH_TOKEN is used, the request still sends the key via the x-api-key header (Anthropic's native auth). However, most proxy endpoints expect Authorization: Bearer <token> instead — this is how Claude Code handles it when ANTHROPIC_AUTH_TOKEN is set (as opposed to ANTHROPIC_API_KEY).

Could build_anthropic_request() be updated to detect which env var was used and switch the auth header accordingly? Something like:

• ANTHROPIC_API_KEY → x-api-key: <key> (current behavior, native Anthropic)
• ANTHROPIC_AUTH_TOKEN → Authorization: Bearer <token> (proxy-compatible, matches Claude Code behavior)

This would align with how Anthropic's own tooling distinguishes between the two variables. See the Claude Code third-party integrations docs for reference.