What Mercury Flux is
Mercury Flux is the AI cost optimization layer from Mercury Technology Solutions. It is the cost-effective update to Mercury Core: a multi-model router that sends every task to the cheapest capable tier (local, cheap, mid, premium), verifies quality with deterministic tests, and ledgers every token. Install it as a Booster Pack on Mercury Core, or run it as a standalone service with no Core license.
What executives should remember
- Flux is the cost-effective update to Mercury Core — same family, cheaper path.
- It also runs standalone. No Core license is required to route, ledger, or verify.
- Typical savings are 60–75% versus an all-premium baseline on mixed-difficulty work.
- Quality is gated by real tests. Cheap models cannot silently degrade output.
- Herdr Core is open source. Flux Cloud and Flux Intel are optional commercial layers.
A cheaper path into Core — or a service on its own.
Mercury Flux is the cost-effective update to Mercury Core. On Core it snaps in as a Booster Pack: same memory fabric, same command center, routed spend. Off Core it is a standalone orchestration service — pip-installable, self-host friendly, no Core license required.
- 1On Core — routing, ledger, and quality gates share OpenClaw memory with your agents.
- 2Standalone — one config file between you and every model you already pay.
- 3Same engine either way. Herdr Core is open source; commercial layers are optional.
You're overpaying for intelligence you don't use.
Premium prices for junior work
Renaming config keys doesn't need a $15/M-token model. Neither does 70% of your workload — but that's what you're paying for it.
Locked to one provider
Every switch means rewrites, retesting, renegotiation. So you never switch — and your provider's pricing team knows it.
A black-box bill
AI spend lands on the credit card with no breakdown by feature, project, or client. Good luck defending it at budget review.
One layer between you and every model.
Flux sits above the providers. Swap any model by editing one config line — adopt next quarter's cheaper model this afternoon, not next sprint.
Triage
A lightweight classifier scores each task's difficulty — trivial to expert — for fractions of a cent.
Route
Tasks flow to the cheapest capable tier: local → cheap → mid → premium. Speculative execution tries cheaper first.
Verify
Deterministic tests run in a sandbox before any model reviews anything. Shaky approvals get a second opinion.
Learn
Outcomes feed back into routing. The engine gets sharper — and cheaper — every week it runs.
# Four lines of config. No rewrites. stages: executor: tier: mid # Flux speculatively tries cheaper first fallback_tier: premium budget: projects: client_a: 500000 # hard cap, enforced before every call
Works with the providers you already pay — Claude, GPT, DeepSeek, Qwen, or your own vLLM/Ollama GPUs.
Measured on your workload, not our slide deck.
Savings depend on workload mix — the Flux dashboard shows yours exactly, from your own runs. Figures are versus an all-premium baseline on mixed-difficulty work.
An asset that compounds, not a bill that grows.
It learns your codebase
The learning router tracks which tier succeeds at which task type in your repos and adjusts automatically. Your routing intelligence never leaves your infrastructure.
Quality is enforced, not hoped for
Real test suites gate every change; low-confidence approvals trigger an automatic second opinion. Cheap models can't quietly degrade output.
It never goes down
Per-provider circuit breakers with automatic failover up the tier ladder. One provider's outage is your non-event.
Built for agencies
Hard per-project budgets mean Client A can never burn Client B's tokens. Every dollar attributed, every run ledgered.
Self-host friendly
Route routine work to your own GPUs at zero marginal cost. Rent what's spiky, own what's predictable.
CFO-ready
Spend per project, per stage, per feature — with savings vs. baseline on one dashboard. AI goes from black-box expense to managed portfolio.
Open at the core. Yours at the edges.
The routing engine is open source — audit it, fork it, self-host it, never pay us a cent. Commercial layers add hosted intelligence.
Herdr Core
Free · forever
- →Full routing engine + 4-tier ladder
- →Learning router + outcome tracking
- →Local ledger and cost dashboard
- →Circuit breakers and failover
Flux Cloud
Per project · pays for itself in one saved premium day
- →Everything in Herdr Core
- →Hosted dashboards and alerting
- →Team budgets and roles
- →Slack / webhook cost alerts
Flux Intel
Custom
- →Everything in Flux Cloud
- →Cross-fleet routing benchmarks from opt-in anonymized data
- →SSO, audit export, SLA
- →Booster Pack install on Mercury Core
Frequently asked questions
Stop paying senior prices for junior work.
five minutes to your first routed run · MIT-licensed core