Platform Capabilities
The complete governance stack
for production AI
Every feature DoCoreAI ships is built on the same principle — govern AI cost and risk inside your runtime, with zero prompt exposure, zero gateway dependency, and zero code changes to your existing stack.
🔒 Privacy & Security
💰 Cost Governance
📈 Scale & Reliability
📊 Analytics
⚙️ Architecture
🛣️ Roadmap
Privacy & Security
Governance that starts with trust
Every privacy feature in DoCoreAI is an architectural decision, not a configuration option. These are not features you turn on — they are guarantees built into how the product works.
All Plans
Zero Prompt Storage
Your prompts, completions, and customer payloads never reach DoCoreAI
servers. Only cost metadata is collected — model name, token count,
latency, and cost. This is not a setting. It is the architecture.
There is nothing to breach because there is nothing to retain.
Removes the compliance blocker that prevents proxy-based tools from
deploying in regulated industries. The single claim that opens
doors others cannot.
Paid Plans
PII Detection at Edge
Scans every outgoing prompt for email addresses, SSNs, credit card
numbers, phone numbers, and IP addresses — before the API call is
made. On detection: hard block. Not a warning. Not a redaction.
The LLM call never happens, and no PII value is stored anywhere.
Catches sensitive data at the source, not after it has been logged
downstream. Stronger than post-hoc log scanning — prevents the
incident rather than detecting it after the fact.
All Plans
Local SQLite Storage
All raw telemetry is written to a SQLite database created in your
environment at install time. Core governance — budget control,
prediction, pacing — runs entirely locally. No cloud connectivity
required to operate. Your data stays on your infrastructure.
Pairs with Zero Prompt Storage to make the "no data leaves your
network" claim verifiable by security teams — not just a marketing
statement, but an auditable architectural fact.
All Plans
Auto-Patch LLM SDKs
Activates via a
.pth sidecar injected at install time.
Wraps OpenAI, Anthropic, Gemini, Groq, Bedrock, and Ollama SDKs
automatically. Zero changes to your existing application code.
One pip install — governance starts on the next run.
Removes the primary adoption friction for developer-led evaluation.
Time to first governed call: under 15 minutes.
All Plans
Fail-Open Architecture
DoCoreAI never sits in your LLM request path. If governance
instrumentation fails for any reason, your LLM calls continue
uninterrupted. DoCoreAI cannot cause a production outage — by design.
The product observes; it does not proxy.
Eliminates the "what if your SDK breaks our app?" objection entirely.
Security review question number one — answered before it is asked.
Cost Governance
Stop paying for the ceiling
Most AI spend is waste — tokens reserved by a default
max_tokens that responses never actually need.
DoCoreAI predicts exactly how many tokens each request needs and
sets a tight ceiling before the call is made. Automatically.
Paid Plans
Token Prediction Engine
A LightGBM model trains locally on your actual usage patterns.
After 14 days of real traffic, it predicts the exact token ceiling
each request needs — before the LLM call is made. Sets a tight,
per-request ceiling instead of the provider's bloated default.
Eliminates 40–70% of token waste per call.
The primary ROI driver. The feature that justifies the entire product.
You pay for the ceiling — this shrinks the ceiling on every single call.
Paid Plans
Budget Pacing Engine
Distributes your daily AI budget evenly across 24 hours.
Prevents early-hour budget exhaustion — the 9 AM spike that
leaves your application unresponsive by noon. Budget pacing
runs entirely in-process. No cloud compute. No manual intervention.
Autonomous from day one.
Critical for any SaaS with 24/7 uptime obligations. Closes the
"budget exhausted by noon" failure mode that no amount of monitoring
alone can prevent.
Paid Plans
Peak-Aware Adaptive Pacing
Learns seasonal and cyclical usage patterns from historical telemetry.
Allows temporary over-pace during known peak windows — product launches,
Black Friday, end-of-month processing — while compensating
in off-peak hours to keep the monthly budget intact.
Prevents throttling during your highest-revenue moments without
blowing the monthly cap. A vertical differentiator for e-commerce
and seasonal workloads.
Paid Plans
Budget Tracking & Alerts
Real-time per-request cost tracking by team, feature, and model.
Alerts fire when spend approaches daily caps or when anomalies
are detected — before limits are hit, not after. Finance and ops
teams get AI spend visibility like any other cost centre.
The management layer. Transforms AI spend from a black-box line item
into a monitored, accountable operational metric.
Enterprise
Per-Team Cost Attribution
Break down AI spend by team, product feature, or environment.
Enables chargeback and showback across business units. Know
exactly which team or feature drives the AI bill — not just
the total. Multi-dimensional aggregation across environments
and providers.
Transforms AI spend from a shared cost into an attributable cost
centre. Primary hook for multi-team enterprise pricing.
The feature that justifies the team-level seat model.
Scale & Reliability
Governance that adapts as you grow
A prediction model trained on last month's traffic is wrong about this month's. DoCoreAI detects when your usage patterns shift and retrains automatically — keeping governance accuracy without manual intervention.
Paid Plans
Drift Detection
Monitors whether the prediction model is still accurate against
live traffic. When usage patterns diverge from the trained baseline —
new features, provider changes, seasonal shifts — drift detection
flags it and triggers automatic retraining before governance
accuracy decays.
Justifies the "autonomous" label. Without drift detection, the
Token Prediction Engine silently degrades over time. This is the
self-correcting mechanism that makes governance sustainable.
Paid Plans
Auto-Retraining
The prediction model retrains automatically as usage patterns
shift — using a 30-day learning window, with a ≥5% improvement
threshold before promotion. No manual model management. No
retraining schedules to configure. Governance adapts to your
workload over time, invisibly.
The 30-day training window on real usage data creates meaningful
switching cost. Customers with trained models have a governance
layer calibrated to their specific workload — not a generic baseline.
Paid Plans
A/B Model Testing
Compare cost and quality metrics across LLM providers side by side
using your own live call metadata — no additional API calls, no
synthetic benchmarks. See the dollar impact of switching from
GPT-4 to Claude or Gemini before committing to the change.
Turns model selection from a gut decision into a data-driven financial
decision. Positions DoCoreAI as the governance layer above provider
choice — not tied to any single LLM.
Analytics & Observability
Every stakeholder gets the view they need
Developers need a signal to act on. Engineering managers need a number to report upward. Finance needs a cost centre. DoCoreAI gives all three — from the same metadata, with zero prompt content involved.
Paid Plans
Cloud Dashboard
Central management view of spend curves, per-team attribution, model
comparison, and anomaly detection. Real-time data streaming with
full history on paid tiers. The layer where developer telemetry
becomes business intelligence for stakeholders above the engineer.
Drives dashboard stickiness and multi-stakeholder engagement.
When finance, ops, and engineering all use the same view, DoCoreAI
becomes embedded in the organisation — not just a developer tool.
Paid Plans
Prompt Health Monitoring
Over-generation indicators and output stability signals derived from
token metadata — not prompt content. Identifies when responses are
consistently longer than needed, or when retry rates spike.
Gives developers an actionable signal to shorten responses and
reduce retries without touching business logic.
The post-call complement to the Token Prediction Engine. Pre-call:
predict and set a tight ceiling. Post-call: measure whether
responses are healthy and flag when they are not.
Paid Plans
Developer Time Saved Analytics
Translates reduced token waste and retry reduction into an
engineering-hours-saved estimate. Gives engineering managers a
metric they can report upward — and gives procurement a payback
period they can use to justify the spend in a business case.
Turns a cost number into an ROI narrative. The metric that moves
a budget approval from a developer's Slack message to a finance
team's spreadsheet.
Enterprise
Long-term Audit Trail
Compliance-grade telemetry retention — 1 year base, configurable
to 7+ years for regulated industries. A metadata-only ledger means
a compliant audit trail with zero prompt content risk. Satisfies
SOX, HIPAA, and GDPR audit record requirements without storing
a single prompt.
The "7+ year configurable audit retention" claim is a direct
enterprise procurement checkbox. The metadata-only approach means
regulated industries get the audit trail they need without the
compliance liability they are trying to avoid.
Architecture & Integration
No proxy. No latency. No lock-in.
DoCoreAI is the only AI governance layer that does not sit between your application and your LLM provider. No proxy infrastructure, no added latency, no single point of failure in the request path.
All Plans
Multi-Provider Support
Native SDK wrappers for all six major LLM providers —
OpenAI, Anthropic, Gemini, Groq, AWS Bedrock, and Ollama.
Governance applies uniformly across your full provider stack
without integration changes when you swap providers.
Single install. All providers governed.
DoCoreAI governs above the provider layer — not inside it.
Customers are not locked to any single LLM, and neither is
their governance stack.
All Plans
No-Gateway Architecture
Gateways add latency, create a single point of failure, and route
your prompts through a third-party network. DoCoreAI adds none of
that. It runs inside your Python process via SDK monkey-patching.
Your app calls the LLM directly. DoCoreAI listens, extracts metadata,
and applies governance — without ever touching the request path.
Zero added latency. Zero proxy dependency. Zero prompt data in
transit through DoCoreAI's infrastructure. The structural margin
advantage that scales: our infra cost does not grow with your call volume.
Roadmap
What ships next
These capabilities are in active development for DoCoreAI v2.2+. Enterprise design partners get early access — reach out if any of these are blockers for your deployment.
🛣️
Not yet shipped. The features below are on the
active roadmap and not available in the current v2.1.0 release.
Status updates are posted on
docoreai.com.
In Development
Agentic Workload Support
Call-graph tracking, agent-aware token prediction, and multi-step
budget allocation for agentic AI workloads. Token usage in agent
runs is high-variance and unpredictable — the governance problem
is acute and growing.
Agentic workloads are the fastest-growing LLM cost source.
The primary capability unlock for AI-native companies.
Planned
Multi-Turn, Multi-Agent & RAG Support
Context-window-aware governance, session-level budget tracking,
and retrieval cost modelling for RAG pipelines. Extends governance
from single-shot calls to full AI workflows — covering the complete
range of production AI architectures.
Each architecture type — multi-turn chat, multi-agent orchestration,
RAG — is a distinct governance surface with distinct cost dynamics.
Q3 2026
Multi-Tenant Governance, RBAC & SSO
IAM infrastructure, SAML/SSO integration, role-based policy engine,
and per-tenant data isolation. Without RBAC and SSO, large enterprises
cannot deploy. SOC 2 audit targeted alongside this release.
Enterprise procurement checkbox — required for large-scale deployment.
The capability that unlocks the full enterprise tier. Procurement
cannot approve without it. Targeted Q3 2026.
Planned
Regional Data Residency
Multi-region cloud infrastructure, data routing logic, and
regional compliance documentation for organisations with strict
data localisation requirements — EU-only, India data localisation,
and other regional data sovereignty frameworks.
Enables deployment in markets currently blocked by data sovereignty
requirements. High willingness to pay from the small volume of
enterprises that need it.
Capability by Plan
Which features ship with which plan
Every plan includes the full privacy and security architecture. Paid tiers unlock the cost governance and analytics stack. Enterprise adds team-level attribution, audit retention, and SSO.
| Feature | Category | Free | Startup | Team | Scale | Enterprise |
|---|---|---|---|---|---|---|
| Zero Prompt Storage | Privacy | ✓ | ✓ | ✓ | ✓ | ✓ |
| Local SQLite Storage | Privacy | ✓ | ✓ | ✓ | ✓ | ✓ |
| Auto-Patch LLM SDKs | Privacy | ✓ | ✓ | ✓ | ✓ | ✓ |
| Fail-Open Architecture | Privacy | ✓ | ✓ | ✓ | ✓ | ✓ |
| No-Gateway Architecture | Architecture | ✓ | ✓ | ✓ | ✓ | ✓ |
| Multi-Provider Support (6) | Architecture | ✓ | ✓ | ✓ | ✓ | ✓ |
| PII Detection at Edge | Privacy | — | ✓ | ✓ | ✓ | ✓ |
| Token Prediction Engine | Cost | — | ✓ | ✓ | ✓ | ✓ |
| Budget Pacing Engine | Cost | — | ✓ | ✓ | ✓ | ✓ |
| Budget Tracking & Alerts | Cost | — | ✓ | ✓ | ✓ | ✓ |
| Drift Detection | Scale | — | ✓ | ✓ | ✓ | ✓ |
| Auto-Retraining | Scale | — | ✓ | ✓ | ✓ | ✓ |
| Cloud Dashboard | Analytics | Limited | ✓ | ✓ | ✓ | ✓ |
| Prompt Health Monitoring | Analytics | Limited | ✓ | ✓ | ✓ | ✓ |
| Developer Time Saved Analytics | Analytics | — | ✓ | ✓ | ✓ | ✓ |
| Peak-Aware Adaptive Pacing | Cost | — | — | ✓ | ✓ | ✓ |
| A/B Model Testing | Scale | — | — | ✓ | ✓ | ✓ |
| Per-Team Cost Attribution | Cost | — | — | — | ✓ | ✓ |
| Long-term Audit Trail | Analytics | — | — | — | 1 year | 7+ years |
| Multi-Tenant / RBAC / SSO | Enterprise | — | — | — | — | Q3 2026 |
✓ Included — Not included Yellow = partial or scheduled · See full pricing →
Start governing your
AI costs today
60-day governance pilot. Full feature access. No code changes to your existing stack.