ZIZKA-AI-SL/ZizkaDB

The database must set rules for multi agentic tests and A2A protocol with actionable insiights for enterprises inside dashboard VPN Deployment for Code Blocks

Create agent to agent wedge and and then test the communication parameters, we have tested it , its working fine but i believe there are nuances that need to be tested and verified

Feature: Error Explorer (F3) A single place to see all failures across your agents, grouped into signatures for triage, with one-click Causal Trace / Impact Trace on any instance. Problem Errors are scattered across agents/sessions. There's no way to answer "what's failing right now, how often, and where" without manually paging through logs. Scope Backend: GET /v1/events/errors?agent=&window=&limit= — groups error events (event_type ILIKE '%error%'/'%fail%' OR data ? 'error') by (event_type, data->>'error') signature within a window (24h|7d|30d|all), with count, agents/sessions affected, first/last seen, and sample event ids. Uses idx_events_type; tenant-isolated + agent-scope aware. Frontend: /dashboard/debugging/errors page — window filter, ranked error group cards (signature, counts, affected agents), sample instances each linking to Root cause (Causal Trace) and Impact (Impact Trace). Loading/empty/error states + a clean "no errors" success state. Nav: Error Explorer entry in

Feature: Impact Trace (F2) The downstream mirror of Causal Trace. Causal Trace answers "why did this happen?" by walking parent_event_id up to the origin. Impact Trace answers the opposite — "what did this event cause?" — by walking the causal graph down through children. Problem Today you can only trace upstream (GET /v1/events/{id}/why). When debugging a bad decision or a failing tool call, you can't see its blast radius — the downstream actions it triggered before the problem was caught. This completes the causal picture (why = up, impact = down). Scope Backend: new GET /v1/events/{id}/impact?depth= — recursive WITH RECURSIVE walking forward (JOIN ON e.parent_event_id = it.event_id), bounded by depth, reusing idx_events_parent. Returns a flat node list with per-node depth so the client can render the fan-out tree. Enforces tenant isolation + agent-scope (assert_agent_allowed). Security hardening (G5): also backfills assert_agent_allowed on the existing why endpoint (scoped

One of the core goals of this product is to make our self-hosted SDKs easy to use. Currently, we offer five SDKs: Python, NPM, MCP, LangChain, and CrewAI. Together, they cover the technologies most developers use to build AI agents. Now we need to stress-test every SDK in real production environments on localhost, identify any issues, and improve the developer experience especially the ease of use for anyone building and running AI agents. Let's do it together. Stress-test every SDK, find what can be improved, and contribute some cool PRs.