Pricing
From 95% to 99.5%. The Last Mile of Agent Reliability.
On hard competitive programming, the same agent goes from 85.7% to 100% pass rate. Same model, same tasks.
Super
dynamic + adaptive · all harnesses
Tailored reasoning. The harness rewrites the cognitive operation to fit your specific task. Safety checks stay locked.
- ✓5,000 dynamic calls/month
- ✓1,500 adaptive calls/month
- ✓4 harnesses · 679 cognitive abilities
- ✓Safety locks always active (failure guard, suppression, checkpoint)
- ✓Hosted MCP at api.ejentum.com/mcp
Go
dynamic + adaptive · all harnesses
Dynamic reasoning across all four harnesses, with a starter pool of adaptive calls when you need them.
- ✓1,000 dynamic calls/month
- ✓250 adaptive calls/month
- ✓4 harnesses · 679 cognitive abilities
- ✓Same API surface as Super
Free trial
one month · no card
See whether your agent actually changes behavior under the harness before paying for it.
- ✓1,000 dynamic calls
- ✓Dynamic modes only (no adaptive)
- ✓All four harnesses unlocked
- ✓No payment method required
Running higher volume? Reach out at [email protected].
Frequently Asked
More answers in the full FAQ · Quickstart guide · Use cases
What is the difference between dynamic and adaptive calls?
Dynamic returns the best-matching cognitive operation for your task as-is. It includes a failure guard, a step-by-step approach, a checkpoint, and a list of shortcuts to suppress. Adaptive does the same retrieval, then rewrites the approach to fit the specific shape of your task. The safety pieces (failure guard, suppression, checkpoint) stay the same; the reasoning steps adapt. Adaptive uses more compute per call, so it has its own monthly pool.
Go or Super. Which do I need?
Start with Go. Same API surface, same four harnesses, 1,000 dynamic + 250 adaptive calls/month. If you actually run out of adaptive calls or you want production volume, Super gives you 5,000 dynamic + 1,500 adaptive at €25/month. You can upgrade from your dashboard at any time.
Which harness should I use when?
Reasoning when your agent makes multi-step decisions, analyzes causality, or plans. Code when your agent writes or debugs code. Anti-deception when your agent talks to users and could flatter or fabricate. Memory when your agent reads people, signals, or context drift across long conversations. Most agents use one or two; you can call any harness from any tier.
What happens when I hit my call limit?
You receive a 429 response with a clear message and a link to upgrade. Pools reset on a 30-day rolling cycle from the start of your subscription. Burst rate is 100 requests per minute across all paid tiers. If you exhaust the adaptive pool, calls in adaptive modes silently fall back to their dynamic equivalent with a header that flags the fallback, so your agent keeps working.
Is there a free trial?
Yes. One month, 1,000 dynamic calls, no credit card required. Create an account, generate a key from your dashboard, and start calling the API immediately. Same endpoint, same abilities, same response format as paid plans. Adaptive modes are not included in the free trial.
What uptime do you guarantee?
We target 99.9% availability on a global edge network with automatic failover. If the API is unreachable, your agent continues on native LLM reasoning. The harness is a reasoning enhancement layer, not a hard dependency in your execution chain.
How do I measure the impact?
Run 50 to 100 representative tasks before and after integrating. Compare task completion rate, reasoning chain quality (does the agent self-monitor, verify, consider alternatives?), and token cost per task. Full evaluation methodology at /docs/evaluate.
Does it work with any LLM?
The abilities are model-agnostic. They are structured reasoning injections, not model-specific prompts. The cognitive operations (suppression, amplification, falsification checkpoints) work with any instruction-following model. We test across major model families.
What exactly does the API return?
A JSON object containing a pre-rendered cognitive operation. Each one includes: a failure guard (the pattern to avoid), a step-by-step approach, a reasoning structure (a graph of steps and decision points), a target pattern (what correct reasoning looks like), a list of shortcuts to suppress, and a falsification check. You inject this into your agent's context before task execution.
Can I cancel anytime?
Yes. Cancel anytime from your dashboard, which routes to the Stripe billing portal. Your access stays active until the end of the current billing period. No clawback, no exit fees.
Are you running a research project or producing technical content?
Email [email protected]. We sometimes provide extended access in exchange for public work (benchmarks, reports, integrations). Not a free tier, not a structured program. Direct conversation, case-by-case.