How to Configure Custom Model Pricing for Cost Tracking in Agentsview

Agentsview allows you to override default model pricing by adding a custom_model_pricing section to your configuration file, which is then merged into the model_pricing table and used for all cost calculations.

Agentsview is an open-source observability tool for AI agents that tracks session costs by multiplying token counts by per-model rates. While the system includes fallback pricing and LiteLLM integration, you can configure custom model pricing to accurately track costs for proprietary, self-hosted, or fine-tuned models. This configuration completely overrides the default rates stored in the model_pricing table.

Understanding the Pricing Hierarchy

Agentsview resolves model pricing through three distinct sources, applied in order of precedence:

1. Hard-coded fallback pricing – defined in internal/pricing/fallback.go and used only when no other source exists.
2. LiteLLM pricing – downloaded at runtime by pricing.FetchLiteLLMPricing and stored in the model_pricing table.
3. User-supplied custom pricing – read from your configuration file and written into the same model_pricing table, overriding any previous entries.

Only the third source allows you to specify exact costs for any model identifier, including those not covered by the public LiteLLM database.

Configuring Custom Model Pricing

Configuration Structure

The configuration system defines the CustomModelPricing field in internal/config/config.go as a map of model patterns to rate structures:

type Config struct {
    // ... other fields ...
    CustomModelPricing map[string]CustomModelRate `json:"custom_model_pricing,omitempty" toml:"custom_model_pricing"`
}

The CustomModelRate struct (also in config.go) supports four distinct rate fields:

type CustomModelRate struct {
    Input         float64 `json:"input" toml:"input"`
    Output        float64 `json:"output" toml:"output"`
    CacheCreation float64 `json:"cache_creation,omitempty" toml:"cache_creation"`
    CacheRead     float64 `json:"cache_read,omitempty" toml:"cache_read"`
}

All values represent cost per million tokens, consistent with the pricing conventions used throughout Agentsview.

File Location and Format

The configuration file resides in your data directory, which defaults to ~/.agentsview (controlled by the $AGENTSVIEW_DATA_DIR environment variable). When the configuration loads (Config.loadFile at lines 563-564 of internal/config/config.go), it parses the custom_model_pricing block directly into the struct.

Create or edit config.toml to add your custom rates:

# ~/.agentsview/config.toml

custom_model_pricing = {
  "gpt-4o-mini" = { input = 0.15, output = 0.60 },
  "claude-3-5-sonnet-20240620" = {
    input = 0.30,
    output = 1.50,
    cache_creation = 0.020,
    cache_read = 0.010
  },
  "anthropic/claude-*" = { input = 0.12, output = 0.40 }
}

The model pattern key supports glob-style wildcards (e.g., anthropic/claude-*) to apply rates to entire model families. Leading underscores are reserved for internal entries and ignored during matching.

How Custom Rates Are Applied

After parsing the configuration, Agentsview commands invoke database.SetCustomPricing to persist your custom rates. This function, implemented in internal/db/pricing.go, writes each entry into the model_pricing table, overriding any existing rows for the same model_pattern.

The injection occurs in three primary command paths:

• CLI usage command (cmd/agentsview/usage.go, lines 147-150)
• PostgreSQL push (cmd/agentsview/pg.go, lines 234-235)
• DuckDB sync (cmd/agentsview/duckdb.go, lines 211-212)

Once stored, the Postgres-specific implementation in internal/postgres/pricing.go loads the combined pricing map via store.loadPricingMap and applies your custom overrides through store.applyCustomPricing during cost calculations.

Verifying Your Configuration

To confirm that your custom pricing is active, query the model_pricing table directly:

SELECT model_pattern, input_per_mtok, output_per_mtok,
       cache_creation_per_mtok, cache_read_per_mtok
FROM model_pricing;

Rows matching your configured patterns should display the exact values you specified in config.toml. If you synchronized to PostgreSQL, you can also verify the rates in the replica database.

Summary

• Agentsview supports three pricing sources with user-defined custom rates taking highest precedence.
• Define rates in ~/.agentsview/config.toml using the custom_model_pricing map with cost per million tokens.
• Use glob patterns in model identifiers to apply rates to entire model families.
• Rates are persisted via database.SetCustomPricing and override the model_pricing table entries used by the usage, PostgreSQL push, and DuckDB sync commands.
• Restart Agentsview after modifying configuration to trigger the reload at Config.loadFile (lines 563-564).

Frequently Asked Questions

What is the priority of custom pricing versus LiteLLM pricing?

Custom pricing always takes precedence. When database.SetCustomPricing executes, it writes your configuration values into the model_pricing table, overwriting any existing rows for matching model_pattern entries—including those downloaded from LiteLLM via pricing.FetchLiteLLMPricing.

Can I use wildcards for model patterns?

Yes. The model pattern key supports glob-style wildcards (e.g., anthropic/claude-* or gpt-4*). This allows you to apply uniform pricing to entire model families without listing each variant individually. Note that patterns beginning with an underscore are reserved for internal use and ignored.

What units should I use for pricing rates?

All rate fields (input, output, cache_creation, cache_read) expect values in USD per million tokens. This aligns with standard API pricing conventions and ensures consistency with the fallback pricing defined in internal/pricing/fallback.go.

Do I need to restart Agentsview after changing pricing?

Yes. The configuration file is read during startup via Config.loadFile in internal/config/config.go. After saving changes to config.toml, restart the Agentsview server or re-run your CLI command (usage, pg push, or duckdb sync) to trigger database.SetCustomPricing and update the model_pricing table.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}