These are proposed improvements, not shipped capabilities or promises. Rank by buyer outcome and a repeatable acceptance check.
P1: one public task-to-verdict example
Add a labelled public fixture, expected output, trace and acceptance record. A misleading source must be rejected and unknown fields preserved. See the Hotido/FinEdg patterns.
P1: operator onboarding and memory correction
Show installation prerequisites, channel health and the path from a wrong remembered fact to correction or expiry. Study OpenClaw/Hermes UX; validate a TVIJO-specific prototype before claiming parity.
P2: visual workflow and recovery
Prototype a visible source-to-review flow inspired by n8n and explicit state inspired by LangGraph. Recovery must preserve identifiers and avoid duplicate alerts. Document any chosen dependency and license; no integration is implied by this comparison.
P2: measure marginal value of the swarm
Run comparable1/3/6-worker arms only after fixtures and budgets exist. Measure accepted results, cost, defects and interventions. Do not use agent count as a quality score.
| Solution & documented option | Delivery / evidence | TVIJO: current scope and useful gap | Proposed domain use |
|---|---|---|---|
| OpenClaw Channel gateway, agent routing and memory | Documented personal-agent runtime Primary documentation; not installed or benchmarked here.
| TVIJO has operator/Telegram components; a comparable packaged multi-channel onboarding and channel-health experience is not established in this review. | Hotido → Offer a source-linked tender alert in a messaging channel. FinEdg → Deliver a read-only research digest with explicit data age. |
| Hermes Agent Persistent memory and reusable skills | Documented assistant runtime Primary documentation; not installed or benchmarked here.
| TVIJO already uses skills and persistent project records. A buyer-facing memory correction, expiry and supersession flow still needs explicit evidence; skills are not absent. | Hotido → Reuse a supplier-checking procedure; correct an outdated identity note. FinEdg → Reuse a research checklist with instrument units and source dates. |
| OpenAI Agents SDK Agent handoffs, traces and evaluation datasets | SDK; hosted evaluation is a separate dependency Primary documentation; not installed or benchmarked here.
| TVIJO has criteria/evidence/review components. A public end-to-end dataset-to-grade example is a packaging target, not proof those mechanisms are missing. | Hotido → Evaluate a tender relevance filter on labelled public examples. FinEdg → Check research answers against sources and numerical references. |
| OpenAI Codex Repository tools, resumable sessions, review and subagents | Coding agent; account quota and availability apply Primary documentation; not installed or benchmarked here.
| Codex is already part of the operator workflow. It can be a worker inside the swarm rather than a mutually exclusive replacement for TVIJO. No like-for-like quality benchmark exists here. | Hotido → Develop a source connector with one owner per file set and parser fixtures. FinEdg → Implement and review paper-research tests with no order permissions. |
| LangGraph Explicit graph state, durable execution and human checkpoints | Orchestration runtime; LangSmith is separate Primary documentation; not installed or benchmarked here.
| TVIJO has board/checkpoint code. A general visual graph editor and end-to-end recovery across all workers are not established. Checkpoint presence alone is not a recovery guarantee. | Hotido → Resume a source-normalisation job after interruption without duplicate notices. FinEdg → Resume a recorded-data research pipeline; keep the order path disconnected. |
| CrewAI Role-based crews separated from control flows | Crew/flow framework; tracing and checkpoints documented Primary documentation; not installed or benchmarked here.
| TVIJO has role/task contracts and an executor ladder. The useful gap is a clear public role-output example and deterministic stop/review flow, not inventing roles again. | Hotido → Collector → verifier → digest writer, each with a declared output contract. FinEdg → Researcher → calculation check → critic, with a fixed stopping rule. |
| n8n Visual workflows, AI steps and Telegram node | Workflow platform; hosting and license terms must be checked Primary documentation; not installed or benchmarked here.
| TVIJO has a custom operator interface and scoped Telegram service. A comparable general-purpose drag-and-drop workflow editor is not established here. | Hotido → Show a buyer a visible source → filter → review → alert prototype. FinEdg → Connect a read-only report to a human review queue; no exchange keys. |
- TVIJO AGENTSJUDJE · 2026-10-15
- TVIJO AGENTSORCH · 2026-10-15
- TVIJO operator guide · 2026-10-15
- TVIJO dated implementation evidence · 2026-10-15