Rizo: BYOK AI coding assistant for VS Code

Providers

Pick a company, then relax

A new chat opens with a one-time choice: Claude, Gemini, OpenAI, DeepSeek, Kimi, or Free. That company is locked in for the whole chat — the in-chat switcher only ever offers its own variants, never a different one. One tool-calling convention, one system-prompt format, one context window, for the life of the conversation. Want a different company? Start a new chat.

Provider	Starts on	Also switch to	Also switch to
Claude	Haiku 4.5	Sonnet 5	Opus 5
Gemini	Flash Lite	Flash	Pro
OpenAI	GPT-5 Nano	GPT-5 Mini	GPT-5
DeepSeek	V3.2 Exp	V3.2	R1
Kimi	K2.5	K2 Thinking	K3
Free	Auto-routed by default, no cost — or pick a specific free model yourself (NVIDIA, OpenAI, Google, Cohere, Z.ai) from the switcher

Every provider starts a new chat on its cheapest variant. Model IDs and pricing are checked against openrouter.ai/models periodically — see the source for exact figures.

Effort

Same model, different depth

A second pill next to the model switcher, adjustable any time, no restart.

Low Medium High

Which model answers never changes

Effort controls how hard the model you already picked thinks — it maps straight to OpenRouter's unified reasoning.effort field, which each provider translates into its own native reasoning budget. It's not a second router: the company and variant you're on stay exactly the same.

Usage you can see, top-left

Today's token count and this month's estimated cost, running totals across every chat and every provider, reset automatically at midnight and on the 1st. Per-chat totals sit under the composer too — nothing buried in a separate dashboard.

Safety

Nothing happens without your click

Every file write and every shell command stops for approval first. A malicious repo can prompt-inject the model into attempting a bad tool call, but it can't click "Approve" for it.

Diff first
Every write shows a real diffVS Code's native diff view, before a single byte touches disk.
Always asks
Destructive commands never get a bypassgit push --force, reset --hard, branch -D, rm -rf and its long-form/colon-refspec equivalents get an elevated warning every single time. "Always Allow" is deliberately not offered for these.
Scoped
"Always Allow" is per-workspaceTrusting one project never carries over into another. VS Code scopes it for us.

BYOK

Bring your own key

Sign up at openrouter.aiFree account, add credit or use the free-tier models with none.

Paste your key into Rizo, onceStored in your OS keychain via VS Code's own Secret Storage, never written to a file.

OpenRouter bills you directlyFor exactly what you use. Rizo has no server sitting in between and never sees a cent of it.

What this means in practice

No shared account, no per-seat pricing, no markup on top of OpenRouter's rate
Your key never leaves your machine except to call OpenRouter directly
The Free provider needs no credit at all: a real way to try it with $0 at risk
Open source, Apache 2.0: the routing and safety logic above is exactly what's in the repo, not a claim to take on faith

Also

The rest of it

Named threads

Separate conversations, each with its own history and its own locked-in provider. Auto-named from the first message, renameable, deletable — with a confirm dialog, since there's no undo.

Cost you can see

Per-chat token/cost totals under the composer, plus a running today/this-month readout at the top of the panel across every chat and provider — not buried in a dashboard.

Terminal & git, agentically

Runs commands in your workspace root with a 60s timeout and a capped output buffer: bounded, not a runaway loop.

Open source

Apache License 2.0. Fork it, read the routing logic, send a PR. Nothing about how it decides where your money goes is hidden.

Pick a provider. Stay in it. Dial the cost yourself.