Why not a generic agent builder
You can build a crew in a weekend. Getting it past your risk committee is the other eleven months.
Generic agent frameworks and workflow tools are good products. This page is about which category of problem each one is actually shaped for, and where the boundary sits.
We compare capability categories, not named vendors. The frameworks are open source; you can verify every row yourself.
An honest capability comparison.
“You build it” is not a criticism. A framework that shipped opinionated governance would be a worse framework. It is simply a description of who does the work and who carries the risk if it is not done.
| Capability | Generic agent framework | Workflow tool + LLM nodes | KrewOS |
|---|---|---|---|
| Orchestrate multiple agents | Yes | Partly | Yes |
| Typed contracts between components | You build it | No | Enforced |
| Evidence with provenance and sufficiency scoring | You build it | No | Built in |
| Abstain below a declared evidence floor | You build it | No | Built in |
| Compile-time proof that untrusted input cannot trigger a side effect | No | No | Built in |
| Immutable releases with dependency lock and evaluation gates | You build it | No | Built in |
| Human approval as a durable workflow state | You build it | Partly | Built in |
| Risk-weighted review sampling with automatic reversion | No | No | Built in |
| Effect verification — idempotency, compensation, no duplicate sends | You build it | Partly | Built in |
| Audit export a regulator would accept | You build it | Partly | Built in |
| Cost per business outcome, fully attributed | You build it | No | Built in |
| Licensed jurisdictional domain assets | No | No | Packs |
| Data residency with a control / execution plane split | You build it | Self-host | Supported |
| Freedom to walk away with your work | Total | Total | Git export & eject |
What you would have to build yourself.
Every row above marked “you build it” is a real system with a real maintenance cost. Enumerated honestly, this is what sits between a working prototype and a crew a risk committee will sign off.
- Manifest schema and a compiler
A canonical machine-readable definition, plus static checks: type resolution across ports, taint analysis from untrusted input to external effects, worst-case budget bounds.
~3 months - Evaluation harness with release gating
Dataset management, scorers, confidence bounds, regression detection, and a promotion gate wired into deployment.
~2 months - Evidence model
Provenance capture, authority grading, sufficiency scoring against a claim, conflict detection, freshness windows and staleness propagation to dependent artefacts.
~3 months - Approval workbench
Queues, routing, SLAs, delegation, immutable correction records, reviewer measurement, and a mobile-tolerable review surface.
~2 months - Secrets and permission gateway
Credential injection outside the prompt, per-component permission scoping, and a guarantee that keys never reach logs, exports or support tooling.
~1.5 months - Durability, idempotency and compensation
Checkpointing, replay semantics, idempotency keys, compensating actions and multi-day wait states that survive a deploy.
~3 months - Cost attribution and outcome accounting
Per-node metering, budget enforcement that binds before spend, and reconciliation to a business outcome unit with zero unattributed spend.
~1.5 months - Audit export and release management
Immutable records, retention policy, scoped export, signing, plus immutable releases with dependency locks and in-flight case continuity.
~2 months
That is roughly eighteen months of a competent team, and then it is yours to maintain forever. It is also not your product. The question is not whether your engineers could build it — they could. It is whether the firm wants to own a governance platform as well as a practice.
When not to use KrewOS.
If your problem is on this list, we will tell you on the call rather than after the pilot.
A single-step task
Summarise this, classify that, extract these fields. Call the model API directly. Governance machinery around one call is overhead with no offsetting benefit.
Deterministic integration
Move a record from A to B on a schedule with no judgement involved. Use a workflow tool. There is no evidence to trace and no decision to attribute.
Consumer-facing chat
High volume, low regulatory exposure, cost per interaction is the binding constraint. Use something cheaper. Our per-run cost carries governance you do not need.
Research prototyping
You want zero structure and maximum freedom to change everything hourly. Use a framework. Typed ports and release gates are friction you have not earned yet.
“We prototyped on a framework in three weeks and it was genuinely good. Then compliance asked for the evidence trail, the approval record and the regression gate, and we were looking at a year of platform work that had nothing to do with our practice.”
Head of Technology, regional professional services group
Book a technical review
90 minutes with a solutions engineer and your architect. Bring your existing prototype. We will map it onto the manifest model, name what is missing for your risk committee, and tell you honestly if the answer is that you do not need us.
Book a technical review