Add Volcengine Ark Agent Plan

providers/volcengine-agent-plan/models/ark-code-latest.toml

@@ -0,0 +1,30 @@
+# Agent Plan console routing alias from
+# https://www.volcengine.com/docs/82379/2373738 (accessed 2026-09-01).
+# The alias follows the model selected in the Agent Plan console; callers do
+# not control that target through this model ID, so no reasoning control is
+# declared. A one-token smoke call returned 200. Subscription usage is metered
+# through Agent Plan AFP rather than per-token API pricing.
+name = "Ark Code Latest"
+description = "Volcengine Agent Plan console-managed routing alias for coding and agent workloads"
+attachment = true
+reasoning = true
+reasoning_options = []
+tool_call = true
+structured_output = true
+temperature = true
+release_date = "2026-08-28"
+last_updated = "2026-09-01"
+open_weights = false
+
+[limit]
+context = 256_000
+output = 32_000
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
+
+[cost]
+input = 0
+output = 0
+cache_read = 0

providers/volcengine-agent-plan/models/deepseek-v4-flash.toml

@@ -0,0 +1,17 @@
+# Toggle: thinking.type = enabled|disabled on POST /api/plan/v3/chat/completions
+# Agent Plan model from https://www.volcengine.com/docs/82379/2373738
+# (accessed 2026-09-01). Subscription tier: no per-token price. The Agent Plan
+# gateway exposes the same toggle-only control as the Coding Plan peer; graded
+# effort values are not declared as distinct controls.
+base_model = "deepseek/deepseek-v4-flash"
+
+[[reasoning_options]]
+type = "toggle"
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 0
+output = 0
+cache_read = 0

providers/volcengine-agent-plan/models/deepseek-v4-pro.toml

@@ -0,0 +1,17 @@
+# Toggle: thinking.type = enabled|disabled on POST /api/plan/v3/chat/completions
+# Agent Plan model from https://www.volcengine.com/docs/82379/2373738
+# (accessed 2026-09-01). Subscription tier: no per-token price. The Agent Plan
+# gateway exposes the same toggle-only control as the Coding Plan peer; graded
+# effort values are not declared as distinct controls.
+base_model = "deepseek/deepseek-v4-pro"
+
+[[reasoning_options]]
+type = "toggle"
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 0
+output = 0
+cache_read = 0

providers/volcengine-agent-plan/models/doubao-seed-2.0-lite.toml

@@ -0,0 +1,16 @@
+# Agent Plan model from https://www.volcengine.com/docs/82379/2373738
+# (accessed 2026-09-01). Subscription tier: no per-token price. Controls match
+# the Coding Plan peer on the same Volcengine subscription gateway.
+base_model = "bytedance-seed/seed-2.0-lite"
+
+[[reasoning_options]]
+type = "effort"
+values = ["none", "low", "high"]
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 0
+output = 0
+cache_read = 0

providers/volcengine-agent-plan/models/doubao-seed-2.0-mini.toml

@@ -0,0 +1,16 @@
+# Agent Plan model from https://www.volcengine.com/docs/82379/2373738
+# (accessed 2026-09-01). Subscription tier: no per-token price. Controls match
+# the adjacent Seed 2.0 Lite peer on this gateway.
+base_model = "bytedance-seed/seed-2.0-mini"
+
+[[reasoning_options]]
+type = "effort"
+values = ["none", "low", "high"]
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 0
+output = 0
+cache_read = 0

providers/volcengine-agent-plan/models/doubao-seed-2.1-turbo.toml

@@ -0,0 +1,16 @@
+# Agent Plan model from https://www.volcengine.com/docs/82379/2373738
+# (accessed 2026-09-01). Subscription tier: no per-token price. Controls match
+# the Coding Plan peer on the same Volcengine subscription gateway.
+base_model = "bytedance-seed/seed-2.1-turbo"
+
+[[reasoning_options]]
+type = "effort"
+values = ["none", "low", "high"]
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 0
+output = 0
+cache_read = 0

providers/volcengine-agent-plan/models/doubao-seed-evolving.toml

@@ -0,0 +1,16 @@
+# Agent Plan model from https://www.volcengine.com/docs/82379/2373738
+# (accessed 2026-09-01). Subscription tier: no per-token price. Controls match
+# the Coding Plan peer on the same Volcengine subscription gateway.
+base_model = "bytedance-seed/seed-evolving"
+
+[[reasoning_options]]
+type = "effort"
+values = ["none", "low", "high"]
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 0
+output = 0
+cache_read = 0

providers/volcengine-agent-plan/models/glm-5.3-flash.toml

@@ -0,0 +1,17 @@
+# Agent Plan model from https://www.volcengine.com/docs/82379/2373738
+# (accessed 2026-09-01). Subscription tier: no per-token price. The base model
+# always reasons. POST /api/plan/v3/chat/completions accepts
+# reasoning_effort = low|high|max; all three returned HTTP 200 in smoke checks.
+base_model = "zhipuai/glm-5.3-flash"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "high", "max"]
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 0
+output = 0
+cache_read = 0

providers/volcengine-agent-plan/models/glm-5.3.toml

@@ -0,0 +1,16 @@
+# Agent Plan model from https://www.volcengine.com/docs/82379/2373738
+# (accessed 2026-09-01). Subscription tier: no per-token price. Controls match
+# the Coding Plan peer on the same Volcengine subscription gateway.
+base_model = "zhipuai/glm-5.3"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "high", "max"]
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 0
+output = 0
+cache_read = 0

providers/volcengine-agent-plan/models/kimi-k2.7-code.toml

@@ -0,0 +1,15 @@
+# Agent Plan model from https://www.volcengine.com/docs/82379/2373738
+# (accessed 2026-09-01). Subscription tier: no per-token price. Thinking is
+# always on and the gateway has no effective caller control, matching the
+# Coding Plan peer.
+base_model = "moonshotai/kimi-k2.7-code"
+
+reasoning_options = []
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 0
+output = 0
+cache_read = 0

providers/volcengine-agent-plan/models/minimax-m3.toml

@@ -0,0 +1,16 @@
+# Toggle: thinking.type = enabled|disabled on POST /api/plan/v3/chat/completions
+# Agent Plan model from https://www.volcengine.com/docs/82379/2373738
+# (accessed 2026-09-01). Subscription tier: no per-token price. Controls match
+# the Coding Plan peer on the same Volcengine subscription gateway.
+base_model = "minimax/MiniMax-M3"
+
+[[reasoning_options]]
+type = "toggle"
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 0
+output = 0
+cache_read = 0

providers/volcengine-agent-plan/provider.toml

@@ -0,0 +1,17 @@
+name = "Volcengine Ark Agent Plan"
+env = ["ARK_AGENT_PLAN_API_KEY"]
+npm = "@ai-sdk/openai-compatible"
+# Agent Plan is a separate subscription tier with its own dedicated API keys
+# and base URL. Generic Ark API keys can authenticate against /api/v3, but are
+# rejected by /api/plan/v3; Coding Plan keys likewise belong to a different
+# subscription surface. Keep this provider distinct so calls cannot bill the
+# wrong plan or hide an entitlement error as an ordinary model failure.
+# The short model IDs below are what the Agent Plan console exposes.
+# `ark-code-latest` is the console-managed routing alias; direct model IDs are
+# also accepted. OpenAI Chat uses POST /api/plan/v3/chat/completions and the
+# Anthropic-compatible API uses POST /api/plan/v1/messages.
+# Sources (accessed 2026-09-01):
+# https://www.volcengine.com/docs/82379/2373738
+# https://console.volcengine.com/ark/region:cn-beijing/subscription/agent-plan
+doc = "https://www.volcengine.com/docs/82379/2373738"
+api = "https://ark.cn-beijing.volces.com/api/plan/v3"