sendnrw/netbird
2
0
Fork 0
mirror of https://github.com/netbirdio/netbird.git synced 2026-09-21 06:09:07 +02:00
Code Issues Packages Projects Releases Wiki Activity

Add Kimi (Moonshot AI) provider to Agent Network catalog

Browse Source 
Adds a first-class kimi_api catalog entry so operators can govern Kimi K3
and K2 Thinking traffic through the Agent Network proxy instead of riding
the generic custom/vllm entry.

ParserID is left empty so the proxy's URL sniffer dispatches both body
shapes Moonshot serves on the same host and key: the OpenAI-compatible
/v1/chat/completions endpoint and the Anthropic-compatible
/anthropic/v1/messages endpoint that Moonshot's official Claude Code
setup uses (same pattern as the Bifrost gateway entry).

Pricing defaults cover both parser surfaces, including the kimi-k3[1m]
model id some Claude Code guides use for the 1M-context alias, so cost
metering doesn't silently skip those requests.

The dashboard's PROVIDER_CATALOG needs a matching entry (separate repo).
This commit is contained in:
mlsmaycon
2026-07-21 04:04:01 +00:00
parent 51f17bf919
commit 4e44210c16
2 changed files with 69 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
management/internals/modules/agentnetwork/catalog/catalog.go
+39
View File
@@ -420,6 +420,45 @@ var providers = []Provider{
{ID: "mistral-embed", Label: "Mistral Embed", InputPer1k: 0.0001, OutputPer1k: 0, ContextWindow: 8192},
},
},
{
ID: "kimi_api",
Kind: KindProvider,
Name: "Kimi (Moonshot AI) API",
Description: "Kimi K3 / K2 models via the Moonshot AI platform",
DefaultHost: "api.moonshot.ai",
AuthHeaderName: "Authorization",
AuthHeaderTemplate: "Bearer ${API_KEY}",
DefaultContentType: "application/json",
BrandColor: "#1A1A2E",
// ParserID empty on purpose: Moonshot serves two body shapes on
// the same host and key, and the proxy's URL sniffer dispatches
// both (same pattern as Bifrost). /v1/chat/completions matches
// OpenAIParser; the Anthropic-compatible endpoint the official
// Claude Code guide uses (/anthropic/v1/messages) contains
// "/v1/messages" and matches AnthropicParser. Pinning "openai"
// here would misparse the Claude Code path — the primary way
// teams consume Kimi for coding today. Both endpoints accept the
// same Moonshot key via Authorization: Bearer (Claude Code's
// ANTHROPIC_AUTH_TOKEN rides that header too).
//
// api.moonshot.ai is the international platform; mainland-China
// accounts live on api.moonshot.cn with separate billing —
// operators there override the host on the provider record. The
// kimi.com subscription coding endpoint (api.kimi.com/coding,
// model id "k3") is account-bound seat licensing rather than a
// meterable platform key, so it's deliberately not the default.
ParserID: "",
// Pricing per Moonshot's platform rates at K3 launch (July 2026):
// K3 $3/$15 per MTok with $0.30 cached input, flat across the
// 1M-token window; K2 Thinking $0.60/$2.50 with $0.15 cache hits.
// Kimi K3 has a single always-on-reasoning SKU — no mini/turbo
// variants at launch. The consumer app's "K3 Swarm Max" mode is
// not an API SKU, so it doesn't appear here.
Models: []Model{
{ID: "kimi-k3", Label: "Kimi K3", InputPer1k: 0.003, OutputPer1k: 0.015, ContextWindow: 1000000},
{ID: "kimi-k2-thinking", Label: "Kimi K2 Thinking", InputPer1k: 0.0006, OutputPer1k: 0.0025, ContextWindow: 262144},
},
},
{
ID: "litellm_proxy",
Kind: KindGateway,