Moving agent topology into a database makes instructions, tool associations, and workflow routing editable as configuration rather than requiring a code change for every adjustment. In CerebrumKit, Islomkhon Nizomkhonov describes that trade as useful when domain experts need to update policy language directly, but less suitable when orchestration needs complex programmatic control flow or tools need stronger isolation.
In his September 20, 2026 article, Islomkhon Nizomkhonov explains CerebrumKit’s database-configured approach through a practical request: “the agent should answer customer questions about their orders.” Here, “topology” means which agents exist, what instructions and tools they have, and the order or grouping in which they run. The author’s account is a description of his implementation, not an independent audit or benchmark.
As an Amazon Associate I earn from qualifying purchases.
What moves from code into the database?
CerebrumKit stores configuration across records and a workflow JSON graph. The author maps the main pieces as follows:
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstall| Configuration | Where CerebrumKit stores it |
|---|---|
| Tools | tools |
| Skills and their tool associations | skills and skill_tool |
| Agents and their skill associations | agents and agent_skill |
| Workflow routing | projects.workflow |
| Tools that provide context before a message | agent_context_tools |
The split is not simply “database instead of Python.” In the author’s design, the tool’s model-facing description and its Python implementation are edited together, while the workflow graph determines how agents and skills are assembled and routed.
#1 Best Overall
Why put agent configuration in a database?
The main benefit is that a domain expert can change policy language and configuration without asking a developer to deploy each change. Nizomkhonov frames the value this way: “The valuable sentence in a support agent is not def lookup_order(...). It is ‘never quote a delivery date you have not read from the order’.” The instruction can matter more to support behavior than the function’s implementation, and storing it as editable configuration makes it easier to revise.
That flexibility changes who can participate in maintaining an agent, but it does not make every change safe for every editor. Policy text and routing can be exposed for configuration, while executable tool bodies remain code that warrants controlled review.
What the order-support example shows
The author illustrates a workflow with two agents. The first recommends giving a customer a credit for a late delivery. The second catches that the order is marked delayed, which does not satisfy the stated eligibility rule requiring an order to be shipped or packed.
Recommended Free Tools
Rank #2
This example shows how a workflow can have agents check or refine one another’s conclusions. It is an illustrative scenario, not evidence of a measured reliability rate or a guarantee that a second agent will catch mistakes in production.
Where the database-driven approach has limits
Complex routing can become awkward
The author says conditional routing with many branches can be cumbersome in a workflow canvas. His recommendation is to use a library when control flow is genuinely a program. A visual or database-backed graph can make common paths easier to inspect and edit, but it is not automatically the best place to express arbitrary branching logic.
Tool bodies are executable code, not isolated scripts
Nizomkhonov discloses that tool bodies run with full builtins and are not sandboxed. He therefore treats tool authoring as admin-only and says changes should be reviewed like code commits. This is an important boundary: moving topology and instructions into editable records does not provide a security sandbox for Python tools. The article does not describe an independent security review.
Conversation history is not replayed
The author says earlier chat transcripts are not replayed into the prompt. That means readers should not assume a conversation automatically carries its full prior transcript into each model call. The article separately identifies agent_context_tools for pre-message context, but does not establish that this mechanism is equivalent to transcript replay or persistent memory.
Process-local state constrains deployment
The described deployment uses one Uvicorn worker because its websocket registry and in-flight task state live in process memory. That is a deployment constraint of the implementation described, not a universal requirement for database-configured agents. Systems that need multiple workers would need an approach to shared or coordinated state beyond the process-local arrangement described in the article.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How to decide whether this tradeoff fits
Database-configured topology is most compelling when changes to policy and agent configuration are frequent, and the people who understand those policies should be able to edit them without a developer-led deployment. The author’s account suggests weighing these questions before choosing the boundary between configuration and code:
Rank #4
- HP ProLiant DL360 G7 8B Server
- 2x X5650 2.66GHz 12-Cores Total
- 32GB RAM / 8x 146GB 10K 2.5in SAS Hard Drives
- P410 w/ 512MB
- How often should a policy or routing change require a deployment?
- Which people can safely edit instructions, tool associations, and workflow configuration?
- Does the workflow need straightforward routing, or genuinely programmatic conditional control flow?
- How will executable tool code be reviewed, and what isolation does it need?
- How will prior conversation context be supplied if transcripts are not replayed?
- Does process-local websocket and task state fit the intended worker and deployment model?
The article does not provide a systematic comparison with another orchestration framework, nor does it report performance or security benchmarks. Its useful contribution is the concrete account of the tradeoffs in one implementation.
Trying CerebrumKit
Nizomkhonov describes a setup path using Docker Compose, a seed script, and a frontend. The commands he gives are:
Free tools Windows power users keep installed
One-click scans. No signup required.
- Start PostgreSQL with
docker compose up -d. - Run
seed_all.py. - Start the frontend with
npm run dev.
The article says admin and client accounts are seeded from .env. Its estimate that setup takes “roughly two minutes” is the author’s estimate, not an independently measured setup benchmark. The steps are presented as the author’s instructions; they have not been independently verified here.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
