Pricing¶
Pay when you need Advanced & Labs in production, managed deployment, or support behind it.
The wedge
The free package proves the wedge: response-level factual-coherence scoring with governed facts. The paid tiers add production entitlement, deployment help, customer evidence, and the premium Director-Class AI category.
:material-layers-outline: What ships where¶
| Tier | Install / delivery | Price | What ships |
|---|---|---|---|
| Director-Lite | pip install director-ai-lite |
USD 0 | Three-line guard() facade, model-free heuristic default, facts/RAG handoff, optional NLI upgrade |
| Director-AI | pip install director-ai |
USD 0 | Full open-core runtime: SDK guard, 5-tier scoring, REST/gRPC server, framework integrations, evidence packets, opt-in streaming contradiction halt |
| Director-AI Pro | Private self-hosted delivery | CHF or USD 490/yr | Advanced production modules and priority support |
| Director-AI Full | Private self-hosted delivery | CHF or USD 980/yr | Pro plus Labs/Research modules and direct engineering support |
| Director-Class AI | Separate commercial product | Scoped order form | Action-control and evidence boundary; deployment and support are scoped separately |
PyPI promotes the whole ladder but only distributes code packages. Publish and
push users into director-ai-lite for the smallest first run, then upgrade them
to director-ai when they need the full runtime. Director-Class AI is a
separate product and repository, sold through a commercial order rather than
as a Director-AI PyPI wheel.
:material-gift: Director-Lite — free package¶
USD 0. Forever. Including production and closed-source.
The first public adoption path is intentionally small:
- :material-check: Three-line
director_ai_lite.guard()front door - :material-check: response-level guard facade
- :material-check: Model-free heuristic default
- :material-check: Facts/RAG handoff through the canonical Director-AI runtime
- :material-check: Optional
director-ai-lite[nli]upgrade path
This is the free tier we promote hardest on PyPI because it shows the core guardrail value without asking users to understand the full platform.
Support the free tiers ☕
Director-Lite and the open-core runtime are free forever, including in production. If they help you, you can buy us a coffee or sponsor on GitHub — entirely optional, and it keeps the free tiers moving.
:material-package-variant: Director-AI — full open-core runtime¶
The full package remains available:
The Apache-2.0 core is genuinely free, not a crippled trial:
- :material-check: Coherence guardrail engine + 5-tier scoring (rules → embeddings → NLI)
- :material-check: Opt-in contradiction-driven streaming halt
- :material-check: SDK guard, FastAPI middleware, REST/gRPC server
- :material-check: Prompt-injection detection, agent/MCP preflight guard
- :material-check: ONNX + Rust-accelerated compute paths
- :material-check: 6 framework integrations (LangChain, LlamaIndex, LangGraph, Haystack, CrewAI, OpenAI)
- :material-check: Self-hosted, air-gapped, no telemetry, no licence key
No source-disclosure obligation. Use it in a closed-source commercial product without paying anyone. Get started
:material-rocket-launch: Free pilot — try the advanced tier on your data¶
30 days · full advanced features · no credit card
- :material-database-search: Bring your own knowledge base — grounded scoring against your documents
- :material-phone: Weekly 30-min call with the maintainer to review results
- :material-check-decagram: Honest assessment — if it does not fit, we tell you
- :material-lock-open: No lock-in, no card, no strings
:material-currency-usd: Commercial tiers — CHF or USD¶
The Advanced & Labs capabilities are source-available under BUSL-1.1: free to read and evaluate, free in non-production. A commercial licence unlocks them in production and hosted/SaaS deployments, and adds support and SLAs.
| Path | Price | Package | Choose it when |
|---|---|---|---|
| Director-Lite | USD 0 | Free PyPI package and Apache-2.0 adoption path | You want the smallest response-level guard integration |
| Director-AI Pro self-host | CHF or USD 490/yr | Advanced-tier production licence and priority support | You run Director-AI inside your own product or internal platform |
| Director-AI Full self-host | CHF or USD 980/yr | Pro plus Labs/Research modules and direct engineering support | You need the complete research curve in a self-hosted deployment |
| Director-Class AI | Scoped order form | Separate action-control and evidence product; deployment and support are scoped separately | You need controlled agent action, auditable decisions, and procurement terms |
At CHF or USD 490/year, Pro self-host costs less than many single senior incident-response engagements. One prevented fabricated citation, incorrect policy quote, or wrong numerical answer can pay for the licence.
For: developers who need the first working response-level guard integration.
pip install director-ai-lite- three-line
guard()facade - model-free default
- optional NLI/RAG upgrade path
- community/self-support
For: teams shipping governed LLM features to production.
- Unlimited deployments
- Advanced & Labs production entitlement
- Priority support
- 99.5 % support target
- self-hosted and air-gapped deployment path
Recurring annual subscription: CHF or USD 490.
For: teams using the complete Advanced and Labs/Research surface.
- Everything in Pro
- Labs/Research production entitlement
- Direct engineering support
- Recurring annual subscription: CHF or USD 980
For: regulated industries, multi-tenant SaaS, high volume, and premium assurance programmes.
- Dedicated support engineer + private Slack/Teams channel
- Custom SLA (99.9 %+)
- On-prem / air-gapped deployment support
- SOC 2 / HIPAA compliance documentation
- Customer-specific evaluation, threshold tuning, and model-adaptation evidence
- Quarterly architecture reviews + roadmap input
Director-Class AI is not offered through an active self-service checkout. Licence, deployment, and support terms are recorded in a scoped order form.
:material-api: Usage-based API (roadmap; not currently sold)¶
The following volume model is exploratory and is not an offer or active checkout:
| Volume / month | Per 1,000 checks | |
|---|---|---|
| First 5,000 | Free | no card needed |
| 5,001 – 50,000 | USD 2.50 | ~USD 0.0025 / check |
| 50,001 – 500,000 | USD 1.50 | volume discount |
| 500,000+ | Custom | contact us |
Self-hosted deployments use the licence tiers above.
:material-compare: How we compare¶
The measured differences, from the committed competitor benchmark — not marketing claims.
| Director-AI | NeMo Guardrails | GuardrailsAI | SelfCheckGPT | |
|---|---|---|---|---|
| Opt-in streaming contradiction halt | :material-check: | :material-close: | :material-close: | :material-close: |
| GPU NLI latency / pair | ~14.6 ms (ONNX GPU) | LLM-bound | LLM-bound | 5–10 s |
| Local guard benchmark p50 / p95 | 0.124 / 0.200 ms | 0.818 / 1.418 ms config-load + LLM | 0.659 / 0.996 ms parse + LLM | not run in this harness |
| Offline / local | :material-check: | :material-close: | :material-close: | :material-close: |
| Self-hosted core price | USD 0 | free | free | free |
| AggreFact balanced acc. | 75.6 % (0.4 B) | N/A | N/A | N/A |
The local guard benchmark was produced on 2026-06-18 on host aaarthuus
(Ubuntu 24.04.4, Linux 6.17, ASRock H510 Pro BTC+, 11th Gen Intel Core
i5-11600K, 12 logical CPUs) and is stored in
benchmarks/results/competitor_guard_latency.json with
non_isolated_local_regression metadata. The local path runs on a 0.4 B model
on commodity hardware. Response-level scoring is the production gate; streaming
contradiction halt is an opt-in interlock for grounded streaming deployments.
:material-scale-balance: Is it worth it?¶
In a scientific, legal, medical, or financial pipeline, a single fabricated number or citation can invalidate everything downstream. The cost of catching it before it ships — at millisecond-scale NLI cost on a small model — is a rounding error next to the cost of one bad answer reaching a customer, a regulator, or a court. The commercial tiers price the advanced safeguards and the support that stands behind them; the core that does the catching is free.
:material-frequently-asked-questions: Questions¶
What exactly is free?
director-ai-lite is free. The Apache-2.0 core in director-ai is also free
for any use, including closed-source production. Only the BUSL-1.1 Advanced
& Labs tier needs a commercial licence for production.
Will PyPI promote all tiers?
Yes. PyPI should explain the ladder and link to pricing. It distributes the
director-ai-lite and director-ai packages; Director-Class AI is promoted
there as the premium service/category, not as a separate wheel.
Do I need a licence to evaluate?
No. The core is Apache-2.0; the advanced tier is free to read and evaluate under BUSL-1.1. Buy a licence when an advanced capability ships to production.
What happens to BUSL files over time?
Each converts automatically to Apache-2.0 on its change date (the fourth anniversary). Today's advanced tier becomes tomorrow's open core.
For the full legal model, see Licensing.