Cohere

The Cohere provider uses provider key cohere and the HPD-Agent.Providers.Cohere package. ModelName is a Cohere chat model id for chat, or a Cohere embedding model id for embeddings.

The provider package is currently net10.0 only because the underlying TryAGI-generated Cohere SDK targets net10.0.

Set an API key:

export COHERE_API_KEY="..."

Use fluent chat setup first:

using HPD.Agent;
using HPD.Agent.Providers.Cohere;

var agent = await new AgentBuilder()
    .WithCohere(model: "command-r-plus")
    .BuildAsync();

var result = await agent.RunAsync("Write one sentence about Cohere setup.");
Console.WriteLine(result.Text);

Provider construction stays small. Put model behavior on chat defaults or per-run chat config, and use Cohere request options only for Cohere-only request fields:

using HPD.Agent;
using HPD.Agent.Providers.Cohere;

var agent = await new AgentBuilder()
    .WithCohere(model: "command-r-plus")
    .WithChatDefaults(chat =>
    {
        chat.Temperature = 0.2;
        chat.TopP = 0.9;
        chat.MaxOutputTokens = 1024;
        chat.Reasoning = new ReasoningOptions
        {
            Effort = ReasoningEffort.High
        };
    })
    .WithCohereChatRequestOptions(cohere =>
    {
        cohere.StrictTools = true;
        cohere.CitationMode = CohereCitationMode.Accurate;
        cohere.SafetyMode = CohereSafetyMode.Strict;
        cohere.ThinkingTokenBudget = 2048;
        cohere.Documents =
        [
            new CohereChatDocument
            {
                Id = "setup",
                Data = new Dictionary<string, object>
                {
                    ["title"] = "Setup note",
                    ["body"] = "Runtime model behavior belongs on ChatDefaults."
                }
            }
        ];
    })
    .BuildAsync();

The equivalent Clients.Chat shape is:

{
  "Clients": {
    "Chat": {
      "ProviderKey": "cohere",
      "ModelName": "command-r-plus",
      "ChatDefaults": {
        "Temperature": 0.2,
        "TopP": 0.9,
        "MaxOutputTokens": 1024,
        "Reasoning": {
          "Effort": "High"
        },
        "AdditionalProperties": {
          "strict_tools": true,
          "citation_mode": "accurate",
          "safety_mode": "strict",
          "thinking_token_budget": 2048,
          "documents": [
            {
              "id": "setup",
              "data": {
                "title": "Setup note",
                "body": "Runtime model behavior belongs on ChatDefaults."
              }
            }
          ]
        }
      }
    }
  }
}

Embeddings

Cohere also registers the embeddings provider family under the same cohere key:

using HPD.Agent;
using HPD.Agent.Providers.Cohere;

var agent = await new AgentBuilder()
    .WithCohereEmbeddings(model: "embed-english-v3.0")
    .BuildAsync();

The equivalent family config is:

{
  "Clients": {
    "Embeddings": {
      "ProviderKey": "cohere",
      "ModelName": "embed-english-v3.0"
    }
  }
}

Runtime Chat Options

Use ChatRunConfig for shared model-call behavior such as temperature, top-p, top-k, max output tokens, stop sequences, tools, tool mode, seed, and generic reasoning options. Put agent-level defaults in ClientProviderConfig.ChatDefaults or .WithChatDefaults(...); use AgentRunConfig.Chat for per-run or per-session overrides.

The Cohere adapter maps the common MEAI chat options to Cohere Chat v2 request fields:

HPD chat option	Cohere request field
ModelId	model
Temperature	temperature
TopP	p
TopK	k
MaxOutputTokens	max_tokens
Seed	seed
StopSequences	stop_sequences
ResponseFormat	response_format
Tools	tools
ToolMode	tool_choice
Reasoning	thinking

Use CohereChatRequestOptions for Cohere-only request properties:

Option	Cohere key	Meaning
StrictTools	strict_tools	Forces tool calls to follow their tool definitions strictly.
Documents	documents	Relevant source documents that the model can cite while generating an answer.
CitationMode	citation_mode	Citation generation mode. Supported values include enabled, disabled, accurate, fast, and off.
SafetyMode	safety_mode	Safety instruction mode. Supported values include contextual, off, and strict.
Logprobs	logprobs	Includes output token log probabilities in the raw Cohere response.
ThinkingEnabled	thinking_enabled	Overrides whether Cohere thinking is enabled. If unset, generic Reasoning controls this.
ThinkingTokenBudget	thinking_token_budget	Maximum number of tokens the model can use for thinking.
Priority	priority	Request priority. Lower numbers indicate higher priority; Cohere defaults to 0.

CohereChatDocument supports:

Option	Meaning
Id	Optional identifier referenced by Cohere citations.
Text	Plain document text for simple string documents.
Data	Structured document data and metadata.

Provider Options

CohereProviderConfig exposes Cohere-specific embedding defaults for the embeddings family:

Option	Purpose
EmbeddingModelId	Default embedding model id for the embeddings provider family.

For embeddings, WithCohereEmbeddings(...) stores the embedding model in the embeddings family config.

Caveats

Cohere chat and embeddings are available only when the HPD-Agent.Providers.Cohere package is referenced by a net10.0 application.

The current TryAGI Cohere MEAI streaming adapter returns a single final update rather than token-by-token streaming.

Validation checks API key, model, endpoint, and embedding model options. Runtime chat options are applied through ChatDefaults and AgentRunConfig.Chat. Generic reasoning is mapped to Cohere thinking; set ThinkingEnabled or ThinkingTokenBudget only when you need Cohere-specific control. Missing credentials fail at BuildAsync() before a live provider call.

Related Reading

• Provider Setup Overview
• Provider Families
• Provider Keys And Environment Variables