Keep the easytokens off yourAPI bill.
Beaver is a small app that runs on your machine, between your coding agent and every model it uses. Easy requests go to open models on your own machine or the community network. Hard ones go to your frontier provider, with your own key. You see what you saved.
- Runs on your machine
- Claude Code, Codex, OpenCode
- Your keys never leave
- Secrets masked
- Savings you can check
How it works
Point your agent at Beaver
Beaver listens on
localhost:4000. Your agent talks to it as if it were the usual API, so you do not change the agent itself.Beaver masks and routes
API keys, tokens and private keys in the prompt are replaced with placeholders before anything leaves your computer. Then simple rules pick a route: images and long context go to your frontier provider; short, simple work stays on open models.
You see the savings
Every request records what it cost and what it would have cost on your frontier model. The difference is your saving, and you can check it by hand.
Works with your agent
Beaver speaks the three APIs that coding agents use: OpenAI Chat Completions, Anthropic Messages and OpenAI Responses.
Beaver is in development. The commands on the right show the planned setup for v0.1 and may change. Ask for early access.
# Claude Code, Codex or OpenCode beaver setup claude-code beaver setup codex beaver setup opencode # What did you save? beaver stats
Three routes
Beaver picks one route per task and keeps it for the session, so caches stay warm.
- A
Local
- When
- Your own machine can serve the request
- Cost
- Free
- Privacy
- Nothing leaves your machine
- B
Community network
- When
- Easy work and your machine is busy or too small
- Cost
- Free during beta
- Privacy
- The prompt reaches a verified node, with secrets masked
- C
Your frontier provider
- When
- Hard work, images, long context
- Cost
- Billed by your provider, on your own key
- Privacy
- Your key stays on your machine; never sent to Tessovix
Use it today
The community network route is live as the Beaver Network API, an OpenAI-compatible endpoint in free beta. Get a key and change the base URL in OpenCode or any OpenAI SDK.
Base URLhttps://api.tessovix.com/v1
List the models your key can use with GET /v1/models. Model names may change during beta. Read the docs or start the guided setup.
curl https://api.tessovix.com/v1/chat/completions \
-H "Authorization: Bearer $TESSOVIX_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "tessovix-mini",
"messages": [{"role": "user", "content": "Write a hello world in Rust"}],
"stream": true
}'
Lend your idle machine
When your machine is idle, it can serve other people's easy requests and earn points you can spend. Machines without a GPU can take part too.
Nodes join by invitation while the network grows. Write to support@tessovix.com. How points are counted.
- PointsSpent on more usage, queue priority and larger models.
- Not moneyPoints are not sold, not transferable and not exchanged for cash.
- LedgerEvery point is recorded in a ledger protected by a hash chain.
- CountWe publish nodes that are actually active, never just registered.
Privacy
- P1Local first.Requests served on your machine never leave it.
- P2Your keys stay with you.Frontier API keys are stored on your machine only. Tessovix never receives them.
- P3Secrets masked.Keys and tokens in prompts are replaced before a request leaves your computer, and restored in the reply.
- P4Training only with consent.Network data is used for training only if you explicitly agree. Data policy.