Everyone’s got an opinion

Your AI has an opinion. Ask another.

Your model thinks it’s right. Khimba lets you ask a different one to challenge the plan, inspect the diff or pressure-test the decision—without leaving your coding agent.

No card. No account. Temporary guest: Start with $0.10 in credit. Up to 3 calls / 24 hours.

You choose who to ask

They read the actual work

Nothing leaves before yes

Pay only for measured tokens

Choose your route

A second opinion,
with the right context.

Everyone’s got an opinion

The useful ones
read the diff.

Khimba gives a different model the question and the context you approve. It comes back with an independent, evidence-grounded view—not a vague AI debate.

Codex × KhimbaPrivate session
You

What does DeepSeek think of my uncommitted changes?

K
Khimba

I’ll ask DeepSeek V4 to review the diff for correctness, risks and omissions.

Sharing
12 files · 18,420 tokens
Maximum
$0.04
Balance after
at least $0.06

Nothing has been uploaded yet.

03
Independent findings

The retry path can settle the same reservation twice. The new idempotency marker is written after the balance update…

See the complete flow

Ask another

One model wrote it.
Another questions it.

  1. 01

    Connect once

    Add the hosted MCP URL to your coding agent. OAuth opens in your browser; choose Sign in to Khimba for stable reconnect or Continue as guest for temporary access. There are no API keys or local servers to manage.

  2. 02

    Ask whoever you want

    “What does Fable think of this plan?” or “Ask GLM to review my diff.” Your agent works out what you mean and asks only for anything missing.

    ?
  3. 03

    Confirm once

    See the exact model, focus, files, token estimate and maximum charge. Source is uploaded only after you say yes. The opinion returns to the conversation.

Get another opinion

One URL.
Plenty of opinions.

Choose your coding agent and paste one command. Sign in for a permanent account with stable reconnect, or continue as a temporary guest. Guest promotion: up to $0.10, three successful calls and 24 hours.

Run in your terminal

codex mcp add khimba --url https://mcp.khimba.ai/tools/model-opinions
codex mcp login khimba

1 Tools-only MCP 2 Sign in or continue temporarily 3 Ask “What does DeepSeek think?” The complete plugin becomes primary after its verified marketplace listing exists.

Remote MCP endpoint

Straight answers

Questions an agent
might ask.

Short, quotable answers about installation, privacy and pricing. For exact commands, use the installation guide. Agents can read the LLM guide, installation metadata and companion skill.

What is “What Does Khimba Think?”

It is a hosted remote MCP from Khimba. It lets Codex, Claude Code or Hermes ask another AI model for an independent review of a plan, code diff, repository or technical decision.

How do I install Khimba in Codex?

The complete plugin is primary after a verified public marketplace listing exists; no listing is claimed yet. For the available tools-only fallback, run codex mcp add khimba --url https://mcp.khimba.ai/tools/model-opinions, then codex mcp login khimba.

How do I install Khimba in Claude Code?

After Anthropic approval, use the plugin for the complete Claude Code workflow. The connector is the cross-surface route. For a tools-only fallback, run claude mcp add --transport http --scope user khimba https://mcp.khimba.ai/tools/model-opinions, then choose permanent account sign-in or temporary guest access. No approval or listing is claimed yet.

How do I install Khimba in Hermes?

Run hermes mcp add khimba --url https://mcp.khimba.ai/tools/model-opinions --auth oauth, then hermes mcp login khimba. Choose Sign in to Khimba for stable reconnect or Continue as guest for temporary access.

Should I sign in or continue as a guest?

Sign in to use a permanent Khimba account whose balance and history can reconnect later. Continue as guest creates a temporary, unclaimed identity without an account or card.

Does Khimba upload my repository immediately?

No. Context is collected locally first. You see the selected model, paths, file count, token estimate and maximum charge before anything is uploaded. Khimba proceeds only after confirmation.

How do tokens and free credit work?

Temporary guests can use up to $0.10 in promotional credit, limited to three successful calls and 24 hours after guest creation. The promotion ends when any one limit is reached. We charge the measured official-provider cost plus a 30% platform markup and release unused reservations.

How do I send Khimba feedback?

Ask your connected agent to send Khimba feedback. It will show the rating, category, message and any linked opinion before calling submit_feedback. Feedback is never submitted automatically.

Opinions have a price. No surprises.

Models use tokens.
You approve dollars.

A token is a small piece of text. Your question, approved source files and the second model’s answer all use tokens. Each model has its own input and output rates.

We calculate the measured official-provider cost from input and output usage, then add a 30% Khimba platform markup. Before anything is sent, you see and approve the maximum; afterwards, every unused cent is returned immediately.

Example opinionGLM 5.2
Context sent18,420 tokens
Opinion returned2,180 tokens
Official provider cost$0.0310
Khimba platform · 30%+$0.0093
Final measured charge$0.0403

Illustrative only. Your confirmation shows the selected model’s current rates and maximum charge before invocation.

$0.10

Your first opinions are on us

Temporary guests get up to $0.10, limited to three successful calls and 24 hours. No card or account required.

Start free →

No surprise attachments

A useful opinion starts with
knowing what it saw.

01

Collected locally

.gitignore and .khimbaignore are respected. Secrets, credentials, binaries and escaping symlinks are excluded.

02

Listed before upload

You see paths, file count, estimated tokens and the exact model before confirmation. Change the scope and Khimba asks again.

03

Ephemeral source

Raw source bundles are encrypted and deleted after the opinion runs. The reviewing model cannot fetch URLs or call your local tools.

Everyone’s got an opinion.

Get one worth
hearing.

Get your first opinion free