FeldsparOpen the app

Early access: Claude and Gemini available now

One API for every model. The cost of every call.

Feldspar is an AI deployment platform. Send the same request to any model and change one parameter to switch. Every response reports its tokens, latency, and cost.

Start with your own key →Request early access
POST /api/v1/chat/completionscurl
curl $FELDSPAR_URL/api/v1/chat/completions \
  -H "Authorization: Bearer $FELDSPAR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-sonnet-5",
    "messages": [
      { "role": "user", "content": "Summarize this ticket." }
    ]
  }'
Every response includesinput_tokensoutput_tokenscost_usdlatency_ms
Available today

A workspace for the models you already pay for

Bring a Claude or Gemini key. You get chat, multi-agent workflows, and a record of every call, all through the same gateway.

  • https://feldspar.gneisslabs.com/app/

    Chat

    Switch models inside one conversation. Each reply records the tokens sent, the tokens received, and the estimated cost.

    Open chat →
  • https://feldspar.gneisslabs.com/app/workflows/

    Agent workflows

    Connect agents into a graph and run it one step at a time. Branch from any checkpoint. Earlier steps are copied, not called again.

    Build a workflow →
  • https://feldspar.gneisslabs.com/app/metrics/

    Metrics

    Median and 95th-percentile latency, failures, tokens, and cost across your recent calls. Open any call to see the exact prompt and result.

    View metrics →
  • https://feldspar.gneisslabs.com/app/account/

    Your own keys

    Connect a Claude or Gemini key. It is encrypted before storage and never displayed again. You see the last four characters. Your provider bills you directly.

    Connect a key →
Unified API

Same request, different model

The request shape does not change between providers. Switch the model field. Nothing else in your code changes.

Model IDProviderStatus
anthropic/claude-sonnet-5AnthropicAvailable
google/gemini-3.5-flash-liteGoogleAvailable
meta-llama/Llama-3.3-70B-InstructOpen sourcePhase 1
Open models · in development

Three ways to run an open model

You choose the trade-off between cold starts and idle cost. Feldspar rents you a dedicated GPU only when you ask for one.

ModeBehaviorYou pay
SharedPopular models stay loaded on shared GPUs. No cold start.Per token
Serverless dedicatedThe model loads on demand and scales to zero. Cold starts occur.Per GPU second
Always-on dedicatedThe GPU stays on. For steady volume or privacy requirements.GPU cost plus markup

LoRA adapters run against a shared base model, so many adapters share one GPU. Rates are not published yet.

Request path · in development

What happens to a request

Plugins run inside the request pipeline, not next to it. You can see, time, and turn off each step.

  1. GatewayAuth, rate limit, balance
  2. Input guardrailsPlugin
  3. RAGPlugin
  4. ModelOpen or closed
  5. ToolsPlugin
  6. Output guardrailsPlugin

Third-party plugins run in a sandbox and declare their permissions: network access, data access, and cost per call. The dashboard shows the latency and cost each one adds.

Marketplace · in development

Attach it to the deployment, not your app

Most teams rebuild the same parts: a PII filter, a retrieval connector, an eval. Feldspar attaches them to a deployment so they stay out of your application code.

  • Guardrails

    PII filter, prompt-injection filter, content safety, topic limits.

  • Tools

    Web search, database query, code execution, MCP servers.

  • RAG connectors

    Google Drive, Notion, S3, websites, SQL.

  • Evaluators

    Hallucination check, groundedness score, regression tests.

  • Harness templates

    A prompt, tools, guardrails, and RAG, preassembled into one setup.

First-party plugins ship before the marketplace opens to other publishers. Tools use MCP, not a proprietary interface.

Pricing

Three tiers, metered per unit

Inference is metered per token, GPU time per second, plugins per call, and RAG per GB of storage and embeddings.

  • BYOK

    Connect your own provider key and pay the provider directly. Feldspar charges a flat monthly fee for the software layer.

    Flat monthly fee

  • Managed

    Buy prepaid credits. Feldspar supplies model access and meters what you use.

    Prepaid credits

  • Enterprise

    Invoiced billing, a custom contract, and private deployment.

    Invoiced

Rates are not set yet. We will publish them when they are final, not before.

Roadmap

What exists, and what does not

This is the full roadmap, including the parts that are not built yet.

  1. Phase 1Building now

    Done. One API for Claude and Gemini. Keys you bring, encrypted at rest. Chat, agent workflows, and per-call metrics.

    Remaining. 5 to 10 open-source models on serverless GPUs. First-party guardrails and tools. Prepaid credits.

  2. Phase 2Next

    Managed RAG as plugins. Shared GPUs for the most-used models. Custom Hugging Face model IDs. Spend caps.

  3. Phase 3Planned

    Open marketplace: plugin SDK, review process, publisher payouts. Harness templates. Fine-tuning presets.

  4. Phase 4Planned

    Invoiced billing. SOC 2. Private deployment. Team roles.

Try it with the key you already have

Connect a Claude or Gemini key and start a chat. Your provider bills you directly. Saving a key makes no model call.

Open the app →Request early access