Start with pay-as-you-go — 33% of what we save you, deducted from your balance. Or pick a subscription with a savings cap. Every dollar of savings is calculated against what your provider would have charged for the same work.
Higher caps, SOC 2, dedicated capacity, on-premise deployments and custom rule engineering.
Every turn passes through ActiveContext. We log what your provider would have charged for the un-curated history (baseline), then what they actually charged for the curated one (actual). The difference is your savings.
Pay-as-you-go: we deduct 33% of your savings from a pre-funded balance — with auto-recharge, like any token provider. No monthly fee, no cap.
Subscriptions include a monthly savings cap. Once you hit it, you choose: either we pass your full message history straight through to your provider unrestructured on your BYOK keys (no further savings), or you top up at the pay-as-you-go rate of 33%. Either way, team rules keep applying.
No fine print. Pick the plan that matches your workload — change later in one click.
| Features | Pay-as-you-go33% of savings | Starter$20 / mo | Pro$50 / mo | Team$100 / seat / mo |
|---|---|---|---|---|
| Core | ||||
| ActiveContext on all sessions | ✓ | ✓ | ✓ | ✓ |
| Per-agent aggressiveness tuning | ✓ | ✓ | ✓ | ✓ |
| Cache-aware curation | ✓ | ✓ | ✓ | ✓ |
| Custom curator rules | ✓ | ✓ | ✓ | ✓ |
| Advanced rule library | — | — | ✓ | ✓ |
| Billing | ||||
| Monthly subscription fee | None | $20 | $50 | $100 / seat |
| Savings cap | No cap | $100 / mo | $400 / mo | $1,000 / seat / mo |
| Pay-as-you-go rate | 33% | Past cap, optional | Past cap, optional | Past cap, optional |
| Auto-recharge balance | ✓ | ✓ | ✓ | ✓ |
| Limits | ||||
| Seats | 1 | 1 | 1 | Unlimited (3+ min) |
| Sessions per month | Unlimited | Unlimited | Unlimited | Unlimited |
| BYO provider keys | ✓ | ✓ | ✓ | ✓ |
| Team & admin | ||||
| Team-wide rule sets | — | — | — | ✓ |
| Manager analytics dashboard | — | — | — | ✓ |
| Cost per developer reporting | — | — | — | ✓ |
| Audit logs · RBAC | — | — | — | ✓ |
| SSO (GitHub, Google) | — | — | — | ✓ |
| Support | ||||
| Documentation · community | ✓ | ✓ | ✓ | ✓ |
| Email support | — | 48-hour SLA | 24-hour SLA | 4-hour SLA |
| Dedicated Slack channel | — | — | — | ✓ |
| Onboarding session | — | — | — | ✓ |
For every turn, we record two prompt sizes: the un-curated history your agent would have sent, and the curated one we actually sent. Multiply by your provider's rate card, subtract our curator-inference cost, and the difference is your savings.
The math is published per session in your dashboard. You can export it. You can audit it. We'd rather show you the math than spin a marketing number.
You choose. Option one: we stop restructuring and pass your full message history straight through to your provider on your BYOK keys — unrestructured, no further savings, no extra charge from us. Option two: top up at the pay-as-you-go rate (33% of additional savings, deducted from your balance) and keep curating.
Either way, your agents keep running and team rules keep applying so behavior stays consistent across developers. We never block, rate-limit, or get in the way of your BYOK usage.
No monthly fee. You pre-fund a balance (like a token provider), and we deduct 33% of every dollar of savings we deliver. Auto-recharge keeps the balance topped up so your agents never pause.
You only ever pay when we save you money. If we save nothing, you pay nothing.
No — not always. For leading open-source models like DeepSeek V4, Kimi K2.6, and GLM 5.1, we bill you directly for LLM usage. No BYOK setup, no separate provider account. Pre-fund your balance and go.
For closed-source models — Anthropic Claude, OpenAI GPT, Google Gemini, and anything else routed through a paid provider — BYOK is required. ActiveContext sits between your agent and the provider on your keys, your contracts, your invoices.
You can mix both in the same workload: direct-billed for open models, BYOK for closed.
No. Curator inference runs asynchronously between agent turns on small, fast models. Net effect on time-to-first-token is under 50ms in our benchmarks — well under typical network jitter to a frontier provider.
Anthropic (Claude Sonnet, Opus, Haiku), Alibaba Cloud Qwen, OpenAI GPT-5, Google Gemini 2.5 Pro & Flash, Mistral, DeepSeek, xAI Grok, plus AWS Bedrock, GCP Vertex, Together, Fireworks, and any OpenAI-compatible self-hosted endpoint.
Anthropic and Alibaba have native prompt caching, which lets ActiveContext deliver its highest savings. Others work — savings are just a bit lower.
Yes. Monthly plans cancel immediately. Annual plans cancel at the end of the term. No clawback on savings already realized.
No. AJNT does not use your prompts, tool calls, or outputs to train any model. We hold session data only long enough to curate it, then it's discarded per your retention policy (default 30 days, configurable to 0).
Start free. See the math on your own workload before you ever pay us.