Enterprise AI That Passes Your Security Review
Annie deploys sovereign AI orchestration on infrastructure you control — no external API dependencies, no data sovereignty risk, full audit trail. Custom engagements for regulated enterprise, government, and defence.
Built for the Whole Buying Committee
Enterprise AI decisions involve multiple stakeholders. Annie is designed to answer each team's questions before they ask them.
All inference runs on your infrastructure. No data leaves your jurisdiction. Full audit trail on every decision. Air-gappable on request.
Open pipeline roles — twelve today. Bring your own models or integrate frontier models via API. 16-week structured implementation included in every engagement.
Platform licence replaces unpredictable per-token costs. Custom scoping means you pay for what your organisation actually needs — no excess, no surprises.
Annual contract structure. Government and defence procurement frameworks supported. Compliance documentation provided as standard throughout the engagement.
Frontier APIs charge per token. Annie is a platform licence.
The pricing model is a meaningful difference. Frontier API customers pay per token — at enterprise volume, that means costs scale linearly with the work the system is doing, and the bill is the one thing you cannot predict at the start of the year. Annie replaces that with an annual platform licence scoped to your deployment. The difference at scale is the difference between a number you can put in a budget and a number that has to be defended every quarter.
| Volume | Annie (annual) | Frontier API (annual) |
|---|---|---|
| 100M tokens / day | $18K – $73K | $550K – $1.8M |
| 500M tokens / day | $90K – $360K | $2.7M – $9M |
| 1B tokens / day | $180K – $720K | $5.5M – $18M |
Why the gap is structural
Frontier APIs charge per token because every inference is a fresh, monolithic forward pass on a hyperscaler GPU fleet. Annie runs small specialist models on standard inference hardware, routes 80–95% of queries to the most cost-efficient path, and amortises the verification pipeline across a single request rather than a single black-box call.
10–100× lower inference cost is the floor, not the ceiling. Cognition Stream fine-tuning cycles start at $5 per run — replacing million-dollar retraining cycles that frontier-API customers cannot access at all.
How an Engagement Starts
No RFP required to start the conversation. A briefing is a structured technical discussion — not a sales pitch.
We walk through your domain, your infrastructure, your compliance requirements, and where AI fits in your operating model. You ask the technical questions that matter to your team.
We design a deployment configuration matched to your requirements — pipeline roles, model strategy, verification architecture, and implementation approach. Presented to your technical and security teams.
Scoped pricing and contract structure aligned to your procurement framework. Government and defence engagements follow standard procurement pathways. Multi-year terms available.
Engagement Tiers
Every tier runs on infrastructure you control. The tiers differ in how deeply you can customise the models and the verification pipeline. Pricing is scoped to your specific deployment during the briefing process.
Deploy a verified sovereign AI platform with one fine-tuned domain specialist.
- Open pipeline role architecture (twelve roles today)
- Plug in your own models (BYOM)
- Frontier model integration via API
- 1 fine-tuned domain specialist model
- Trained on your proprietary data
- Annie Workforce Foundation Model access
- Standard Judgment Panel
- Fixed domain rubrics
- Quarterly Cognition Stream fine-tuning cycles
- Sovereign on-prem deployment
- Full audit trail and explainability logging
- 99.5% uptime SLA
- 16-week implementation program
- Dedicated technical account manager
Multiple fine-tuned specialists, each trained on domain-specific data — orchestrated and verified by Annie.
- Open pipeline role architecture (twelve roles today)
- Plug in your own models (BYOM)
- Frontier model integration via API
- Up to 6 fine-tuned specialist models
- Each trained on domain-specific proprietary data
- Multi-specialist orchestration and routing
- Annie Workforce Foundation Model access
- Configurable Judgment Panel weights
- Adjustable consensus thresholds per domain
- Cross-domain verification and peer consensus
- Monthly Cognition Stream fine-tuning cycles
- Sovereign on-prem deployment
- Full audit trail and explainability logging
- 99.7% uptime SLA
- 16-week implementation program
- Dedicated technical account manager
- Priority engineering support
Full platform control — retrain the Judgment Panel, configure verification logic, personality, and reasoning chains.
- Open pipeline role architecture (twelve roles today)
- Plug in your own models (BYOM)
- Frontier model integration via API
- Unlimited fine-tuned specialist models
- Each trained on domain-specific proprietary data
- Custom pipeline architecture design
- Annie Workforce Foundation Model access
- Full Judgment Panel retraining
- Custom rubrics, consensus logic, verification architecture
- Personality and reasoning chain configuration
- Weekly Cognition Stream cycles (custom scheduling)
- Custom fine-tuning cadence
- Sovereign on-prem or air-gapped deployment
- PROTECTED workload certification support
- Custom SLA
- Dedicated implementation and engineering team
- Custom integration architecture
Included in Every Tier
The platform foundation is consistent across all tiers. The difference is the depth of model and verification customisation.
Deployed entirely on infrastructure you control. No external API calls. No data leaves your jurisdiction. Fully air-gappable on request.
Every pipeline role slot open — twelve today. Bring your own models, integrate frontier models via API, or use Annie's Workforce Foundation Models. Your choice, your architecture.
Every response scored by independent peer models using domain rubrics before delivery. Hallucination is a managed risk, not an accepted one.
Every inference, every verification decision, every routing step is logged and auditable. Regulators get the decision trail they require.
A dedicated implementation team runs the 16-week deployment program — from infrastructure setup and model integration through to team onboarding.
Automated off-peak fine-tuning cycles that improve your specialist models on real interaction data. Starting at $5 per run. No manual retraining required.
Need the architecture to defend the choice internally?
If your team needs the Annie platform architecture on paper before a briefing — for a security review, a procurement evaluation, or an internal stakeholder conversation — the Annie Architecture Overview is a 12-page document covering the pipeline, verification layer, deployment model, and security posture. Gated so we can verify the request is from a real evaluation context.
Common Questions
What does "twelve pipeline role slots" mean, and is that a hard limit?
Annie's platform architecture currently defines twelve roles in the inference pipeline — classification, routing, specialist inference, peer review, judgment, verification, and more. Each slot can be filled by a model you provide, a frontier model accessed via API, or an Annie Workforce Foundation Model. Twelve is the number shipping today, not a hard architectural ceiling — the pipeline is designed to add roles as deployments grow. You decide the mix. You own the models you bring.
What is BYOM (Bring Your Own Model)?
You can bring existing models — proprietary weights, fine-tuned open-source models, or domain-specific models your team has already built — and plug them into any open pipeline role slot. Your model runs inside your infrastructure, through Annie's verification pipeline. It never leaves your jurisdiction.
How does frontier model integration work?
If you have an existing relationship with a frontier provider (OpenAI, Anthropic, Google, Mistral) and want to use their models via API for specific pipeline roles, Annie supports that as a custom deployment. The API call originates from within your infrastructure boundary. This allows you to use frontier capability where it makes sense while keeping sensitive roles on sovereign models.
What is the Judgment Panel and who controls it?
The Judgment Panel is Annie's verification layer — a set of independent peer models that score every expert response against domain rubrics before it is returned to the user. In Foundation tier, the Panel uses standard rubrics. In Specialist tier, you can configure consensus weights and thresholds. In Sovereign tier, you can retrain the Panel entirely — custom rubrics, consensus logic, and verification architecture specific to your organisation.
What happens during the 16-week implementation?
Weeks 1–3: Infrastructure setup, security assessment, and architecture design. Weeks 4–8: Model integration, pipeline configuration, and initial fine-tuning on your data. Weeks 9–12: Integration testing, verification calibration, and user acceptance testing. Weeks 13–16: Staged deployment, team training, monitoring setup, and handover. The implementation program is included in all tiers.
What infrastructure do we need?
Annie runs on standard inference GPUs — not training clusters. A Foundation tier deployment typically requires a modest GPU rack (H100 or equivalent). We'll specify the exact infrastructure requirements during the enterprise briefing once we understand your query volume, domain complexity, and number of pipeline roles.
How is the contract structured?
All tiers are annual contracts. Payment terms are negotiated during the briefing process — typically annual upfront or quarterly. Government and defence engagements follow standard procurement frameworks. Multi-year commitments are available with pricing locked for the contract term.
Is Tier 3 relevant for government and defence?
Yes. Tier 3 is designed for PROTECTED workload environments, air-gapped deployments, and organisations that cannot accept any external dependency — including on Annie's own Workforce Foundation Models. Full retraining of the Judgment Panel and personality configuration is particularly relevant for defence and intelligence applications where standard rubrics may not apply.
Why doesn't this page show prices?
Because a number without context creates the wrong conversation. Annie deployments vary significantly by infrastructure scale, domain complexity, number of specialist models, and compliance requirements. A figure that's right for one organisation is misleading for another. We scope every engagement through the briefing process so the commercial proposal reflects your actual requirements — not a tier card average.
Start With a Briefing
A 60-minute structured discussion covering your domain, your infrastructure, and your compliance requirements. No commitment. We'll tell you what's realistic — and what isn't.
Annual contracts · Sovereign infrastructure · Built in Australia