tokevenIntegrations
One wrapper. Four providers. Every call tracked.
Claude (Anthropic SDK)
Drop-in Python wrapper around the Anthropic SDK. Wrap your client once and every Claude API call it makes is tracked. Claude Code and the Claude apps are not captured by the wrapper.
- Zero-config install
- Automatic cost computation
- Cache-aware pricing
- Session tracking
OpenAI (Codex & GPT)
Same privacy-first wrapper for the OpenAI SDK. Track GPT 5.4, GPT 5.5 and the GPT 5.6 family (Sol, Terra, Luna) with the same metadata-only approach.
- Same privacy model
- Unified dashboard
- Model-tier recommendations
- Cross-provider cost comparison
Google Gemini
Wrap the Google GenAI SDK to track Gemini calls. One dashboard for every provider you use.
- Google SDK wrapper
- Unified cost view
- Cross-provider analytics
- Same local-only privacy
xAI (Grok)
Grok runs on the OpenAI SDK against api.x.ai, so there is nothing extra to install. TokevenXAI tags the traffic as xAI so it prices against Grok rates rather than GPT ones.
- No extra SDK
- Per-model cache pricing
- 200K long-context tier priced correctly
- Same local-only privacy
TypeScript SDK
npm install @tokeven/sdk and wrap your Anthropic, OpenAI, Gemini or Grok client. Runs on Node, Vercel and Cloudflare edge, Deno and Bun.
- Same wrap() as Python
- Streaming included
- Edge and serverless ready
- Same local-only privacy
LiteLLM
One callback tracks every LiteLLM completion, whichever provider it routes to. Works on the LiteLLM proxy too.
- litellm.callbacks = [TokevenLogger()]
- Sync, async and streaming
- Proxy supported
- Budget guard on async calls
LangChain
A callback handler for any LangChain chat model. Token counts come from LangChain's standard usage metadata; message text is never read.
- callbacks=[TokevenCallbackHandler()]
- Any provider LangChain supports
- Budget guard before each call
- Same local-only privacy
OpenAI Agents SDK
A trace processor that records every model call an agent makes, beside the SDK's own tracing. Span inputs and outputs are never read.
- add_trace_processor(TokevenTracingProcessor())
- Chat Completions and Responses API
- Ready-made budget hooks
- Same local-only privacy
VS Code Extension
An editor extension is planned but not yet built. Today, the MCP server brings pre-send estimates into Claude Code and other MCP-compatible editors.
- Inline cost display
- Model picker
- Pre-send estimates in the editor
MCP Server
The pre-send advisor as a Model Context Protocol server for Claude Code and any MCP-compatible tool. It prices a call before you send it. It does not track usage - tracking comes from the wrapper or the sidecar.
- tokeven_estimate: cost and cheaper alternatives, locally
- Confidence score per task type
- tokeven_improve: prompt rewrite on your own Anthropic key (paid)
- Install with the tokeven[mcp] extra
n8n
An n8n preset for outbound webhooks (paid plans). Point it at an n8n Webhook trigger and route budget alerts into Slack, Teams, email, or ticketing systems.
- budget.threshold_crossed event at 50/80/100/150% of monthly budget
- HMAC-SHA256 signed deliveries
- Retries with backoff
- Build the rest of the workflow in n8n
Webhooks & REST API
Signed webhooks push budget alerts to Zapier or any HTTPS endpoint (paid plans). The REST API feeds your usage data into your own BI tools, with CSV and JSON export on paid plans.
- Zapier preset or any URL
- Flat JSON payloads
- Retry with backoff
- CSV and JSON export (paid plans)
Start with one provider. Add the rest in minutes.
The free tier includes the wrapper, MCP server, and API access. No credit card required.
Start free