Venice AI Bitcoin Accounting | Opencode Desktop & Clams Setup

This guide sets up Venice AI Bitcoin accounting with Opencode Desktop and Clams. Venice AI is the private model provider, Opencode is the agent shell, Clams is the accounting engine. One thing to be clear about up front: with Venice as your model provider, your prompts and the tool output the agent sees (which can include wallet labels, amounts, addresses, and report contents) are sent to Venice's infrastructure for inference. Venice states that it does not log prompts, train on your data, or share it with third parties. Those are Venice's claims per their documentation, not guarantees Clams can make. What Clams controls is the other side: your accounting data stays on your machine by default, and every wallet connection, cost basis calculation, and report is produced locally.

The Stack

• Clams CLI: the accounting engine. Connects wallets, tracks cost basis, generates reports.
• Clams Agent Skills: teaches the model every Clams command, workflow, and prerequisite.
• Venice AI: privacy-focused model provider. States no prompt logging and no training on user data (per Venice's own documentation), with a choice of open-weight models.
• Opencode Desktop: open-source AI agent. Connects to any OpenAI-compatible endpoint, including Venice.

Prerequisites

Before doing anything in Venice or Opencode, get Clams installed and verified. Without the CLI in place, the agent has nothing to drive.

The commands below run in your terminal. On macOS that's the Terminal app (open Spotlight with Cmd+Space and search "Terminal"). On Linux, any terminal emulator works. On Windows, use WSL or Git Bash. Paste one command at a time and press enter.

Install the Clams CLI:

curl -fsSL https://clams.tech/install.sh | sh

Run the guided set up:

clams init

Paid workflows like report generation require a signed-in Clams account with an active license. Your accounting data stays local by default; sign-in, licensing, billing, and price rates are the hosted pieces, and any sync happens only when you explicitly ask for it.

Install the agent skill globally so any Opencode project picks it up:

clams skills install --global

Then grab Opencode Desktop from opencode.ai/download. Pick the build for your OS and finish the installer before opening Venice in the next step.

Generate a Venice API Key

Log into your Venice account and open the API tab from the left side menu. You'll land on the API Settings screen where you can purchase credits, check usage, and manage keys. Click Generate API Key in the top-right.

Venice API Settings. Click Generate API Key in the top right.

In the modal that opens, give the key a name. We're using it from Opencode, so label it Opencode to keep things obvious if you generate keys for other clients later. Leave Inference selected for permissions, leave the expiration and spend limits at defaults, then click Generate key.

Name the key "Opencode" and leave the defaults in place.

Venice will show the key exactly once. Copy it and store it somewhere safe. A password manager works fine. If you lose it, you'll need to generate a fresh one.

The key is shown once. Copy it before closing the modal.

Connect Venice to Opencode Desktop

Open Opencode Desktop and click the gear icon at the bottom left to open Settings. Choose Providers in the sidebar. You'll see the built-in list (Anthropic, OpenAI, Google, OpenRouter, and so on). Scroll to the bottom and click Show more providers.

Settings → Providers → Show more providers.

Type venice into the search bar at the top of the Connect provider dialog. Venice AI will appear under Other. Click the row to select it.

Search for "venice" and click the Venice AI row.

Paste the API key you copied earlier into the Venice AI API key field and click Continue. Opencode validates the key against Venice and lists Venice as a connected provider.

Paste the Venice API key and click Continue.

Choose a Venice Model

Back on the main Opencode screen, the model picker sits next to the chat input. Click it to open the model list. Your Venice models now appear grouped under Venice AI. Pick the one you want as default.

Select a Venice model. At the time of writing, Mistral Small 4 is a solid default for agent work.

A few notes on picking a model. Venice's lineup changes often, so treat these as accurate at the time of writing and check the model list in your own picker:

• Mistral Small 4 is fast, cheap, and handles tool calls reliably. Good default for everyday accounting queries.
• Qwen 3.6 27B trades a little latency for stronger reasoning if you're asking the agent to plan multi-step workflows.
• MiniMax M2.7 and Kimi K2.5 are heavier models for multi-step workflows that need more planning before the agent starts firing CLI calls.

You can switch models per conversation, so start with a small one and escalate when you hit something it stumbles on.

Load the Clams Skill

In the chat input, type /clams and hit enter. Opencode shows the matching skill in the suggestion strip above the input. Confirm clams and submit. The agent now has the full Clams skill loaded: every CLI command, the prerequisite checks, the expected report flags, the connection types Clams supports. (In Claude Code, /clams loads the same skill. If you're using a different agent, or the suggestion doesn't appear, check your agent's documentation for how it discovers installed skills.)

Type /clams to load the agent skill for Bitcoin accounting.

In our internal skill evals, agents with the skill loaded complete common workflows reliably that they otherwise fumble. That's the difference between an agent that guesses at flag names and one that runs the right command on the first try.

Talk to Your Books

You're set. Try something like:

• "Add a new xpub connection labelled cold-storage. Here is the xpub: [pasted xpub]"
• "What's my total balance across all wallets?"
• "Show me Lightning transactions from my LND node this month."
• "Generate a Q1 2026 capital gains report as a PDF."
• "What's the cost basis on this transaction? [pasted txid]"

The Clams CLI runs locally against your wallet data. Venice handles the natural-language layer without retaining your prompts or training on them. Your accountant gets the numbers. Your books never leave your control.

Private Inference, Local Books

Venice's no-logging stance means your prompts about wallets, balances, and tax events don't pile up in a vendor's training set. Clams' local-first model means the underlying data never leaves your machine to begin with. Together, they give you a Venice AI Bitcoin accounting setup without the usual privacy tradeoff.

For an alternative private-AI stack that runs prompts through hardware-isolated enclaves on a local proxy, see our Maple AI setup guide.

Once the stack is running, ask your agent for a capital-gains or tax report and Clams produces audit-ready numbers from your own wallet data, privately. New to Clams? See how it works for developers.

Any feedback or issues getting set up, email us at support@clams.tech.

Clams Team