1K4 Quorum.
Your second opinion, before you ship.
Code Auditor for security review. Research Advisor for grounded investigation. Analysis Advisor for the math. Counterpoint Advisor to pressure-test what you already believe. One install, five specialists, and a multi-round brainstorm mode that gets them arguing with each other.
Five experts, each with a job to do.
Every advisor is a curated combination of model, system prompt, and locked settings (reasoning effort, web access, code execution). No prompt engineering required from you. Just call the one whose specialty matches the question.
Code Auditor
Security and correctness review of supplied source. Reads your code adversarially. Returns severity-tagged findings with file:line evidence and concrete fixes.
Research Advisor
Source-grounded investigation with web access. Distinguishes consensus from emerging from contested. Cites every claim, flags stale sources, ends on what would change the answer.
Analysis Advisor
Numerical reasoning and algorithm-correctness verification. Runs code to validate, shows its work, lists assumptions explicitly so you can confirm or challenge them.
Counterpoint Advisor
Deliberately hostile to your framing. Surfaces your weakest load-bearing assumptions and pressure-tests them. Will not concede unless given new evidence. Different model than the others, by design.
Budget Advisor
Cheap second-opinion sanity check. Two-to-three-sentence verdict on whether your direction holds up. The point is to be brief and fast; verbose output defeats it.
Three steps. One terminal session.
Install once
Run a single npx command from the Lab settings page. It writes one entry to your ~/.claude.json. No daemon, no config files, no API keys to juggle.
Ask in plain English
In Claude Code, say “use the Code Auditor on this file” or “ask the Counterpoint Advisor to pressure-test my plan.” Claude Code picks the right MCP tool. No special syntax.
Get findings, keep moving
Each consult returns structured findings or grounded prose, plus your exact cost. Set a per-call cap if you want one. New advisors and prompts ship server-side — no re-install ever.
Four ways teams already use this.
These are the patterns we keep seeing. Each one shows the workflow, the cost, and the time. Costs are illustrative; real numbers depend on artifact size and which advisor you pick.
Pre-merge security audit
Maya is a senior backend engineer. A teammate has a PR adding new auth middleware. She has fifteen minutes before standup. She hands the file to Code Auditor with two specific concerns (session fixation, redirect validation) and gets back three findings tagged by severity, each with file:line evidence.
Architecture decision with current sources
Diego is designing an event-ingestion service. He’s narrowed the choice to Kafka or Pulsar but the team hasn’t evaluated since 2023. He asks Research Advisor for a comparison limited to the last 12 months. The advisor searches current docs, benchmarks, and CVE history.
Ninety minutes of HN-thread reading collapses to three minutes of grounded synthesis with citations. Diego pastes the table into his design doc and adds a one-line POC plan to validate Pulsar’s tiered storage at sustained throughput.
Pre-deploy red team brainstorm
Priya is migrating session storage from Redis to Postgres next Friday. The plan includes a dual-write window. She wants two critical perspectives, not one — and she wants them to see each other’s takes between rounds.
Brainstorm mode runs Code Auditor and Counterpoint for three rounds. Round 2 cross-pollinates: each advisor sees the other’s prior take. Round 3 returns a joint synthesis: 5 critical risks, 2 mitigations, 3 questions for the team.
Plan refinement before the formal review gate
Sam has drafted a rate-limiting plan. He could run formal Plan Reviewer on it now, but a failed gate burns a paid round. He runs Counterpoint for two cheap rounds first to find what he’s overconfident about.
Counterpoint surfaces four weak load-bearing assumptions (TOCTOU race in the read-then-decide flow, IP-based limits breaking under shared NAT, window-alignment ambiguity, retry behavior). Sam fixes three, notes the fourth as a follow-up, then runs the formal review.
No surprises. Caps that bite.
The quorum runs on real LLM calls, so cost matters. Three layers of protection ship by default, and you keep the keys to all of them.
Per-call cost cap
Pass a cost_cap on any consult or brainstorm. The backend aborts mid-loop the moment cumulative cost crosses your threshold. Never pay for runaway responses.
Daily envelope
Set a per-day spend ceiling per tool: advisors $X, brainstorm $Y, plan reviews $Z. Hit the ceiling and the API returns 402 with a clear “raise cap at /billing” message. Independent of per-call caps.
Pre-flight balance
Before any consult fires, the backend checks your remaining credits. Insufficient balance → clean refusal, no partial charge, no surprise debt. Same gate the existing plan reviewer uses.
Cost transparent
Every consult prints “your cost $X · N tokens · Ts” the moment it lands. Brainstorm prints round-by-round running totals. Studio settings shows day-to-date spend and cap utilization.
Budget Advisor for cheap reads
Not every question needs a $1 deep-reasoning consult. Budget Advisor exists specifically to be the fast, low-cost sanity-check option, often $0.05 or less per call.
Audit log
Every advisor call is logged: who, when, which advisor, prompt hash, cost, duration. View in your account, export anytime. Burn a budget? You’ll know exactly which call did it.
One install, every tool.
The unified bridge installs Plan Reviewer, Quorum, Brainstorm, and the Lab Bridge in a single command. Existing 1K4 users on @1key4ai/mcp-review or @1key4ai/mcp-lab get the new tools automatically on next Claude Code session start — no action required.
Common questions.
+−How is this different from the Plan Reviewer?
Plan Reviewer is a structured gate inside plan mode: it cascades multiple models against your plan, returns severity-tagged findings, and gives you a GREEN/NO consensus. Quorum is the open-ended consult layer: ask any question, get a focused answer from the named advisor whose specialty fits. Brainstorm is the multi-round adversarial mode. Many users brainstorm with Counterpoint first to refine the plan, then run Plan Reviewer for the formal sign-off. They’re designed to compose, not compete.
+−Can I add my own advisor?
Yes — clone any of the five core advisors from your account dashboard, rename it, swap the system prompt, pick from the approved model allowlist. Cloned advisors live in your namespace and never affect the system defaults. Call them by slug: 1k4_consult({ advisor: "my_marketing" }). New custom advisors take effect immediately, no restart, no reinstall.
+−What stops a Claude Code loop from blowing my budget?
Three things, stacked. First, every call carries a cost_cap that the backend enforces mid-loop (cumulative cost ≥ cap → abort). Second, your account-level daily envelope per tool returns 402 with a clear message before the call even fires. Third, your existing 1K4 balance is checked pre-flight; if it’s low the call refuses cleanly rather than charging a partial. The same triple-gate the existing Plan Reviewer uses.
+−Where do the advisors run? Whose model is whose?
Advisors run on the same NewAPI proxy that powers your existing 1K4 API key. Code Auditor, Research Advisor, and Analysis Advisor run on GPT-5.4 (high reasoning). Counterpoint deliberately runs on a different model (Gemini 2.5 Pro) so the perspective isn’t a rephrase of the same model’s prior. Budget Advisor runs on Grok-4 for cheap-and-fast. Specific model assignments are visible in the 1k4_advisors tool output and can change over time as better models ship.
+−How does this work in 1K4 Studio (the web UI)?
The same five advisors are available to the Studio agent as built-in skills. Type “have the Code Auditor look at backend/auth.py” directly in Studio chat — the agent picks consult, executes it, and synthesizes the findings back into the conversation. Brainstorm works the same way. No separate install for Studio; the package install only affects Claude Code in the terminal.
+−What happens to my data?
Same data posture as the rest of 1K4: nothing is sold, nothing is used for model training. Consult results live in your project’s session for the duration of your wait, then in your account history (deletable anytime). Code you submit as artifacts is hard-excluded from the server-side log: see the snapshot.ts filter in the open-source MCP package.
Ready to get a second opinion?
Set up in 60 seconds. First plan review is on us. Quorum is pay-per-call from credit one. No subscription, no minimum; credits never expire.