Loading ToolBench
Open-source knowledge graph for professionals — auto-extracts entities and relationships from natural conversation via MCP
Kernal has 13 well-named tools with clear verb-noun patterns (kernal_remember, kernal_add_person, kernal_recall). All tools have descriptions (avg ~120 chars, within baseline 34-392 range). Input schemas are present with types and descriptions for most parameters. However, output schemas are NOT documented, critical gap for LLM planning. Several tools lack parameter constraints (enums where appropriate). Error handling is minimal; no recovery guidance. Tool composition is sound (single responsibility), but descriptions could be more explicit about prerequisites and when to call each tool.
List action items and follow-ups. By default shows open items.
List recent interactions — meetings, calls, emails, events.
Create a follow-up or task. Optionally assign to a person by name.
Log an interaction — meeting, call, email, event, or introduction. Links to participants and organizations by name.
Add an organization to the knowledge base. Automatically deduplicates — if the org already exists (fuzzy name match), updates their record with any new info.
Add a person to the knowledge base. Automatically deduplicates — if the person already exists (fuzzy name match), updates their record with any new info instead of creating a duplicate.
Output schemas not documented. LLMs cannot plan downstream tool calls or extract required fields (e.g., what does kernal_context return? Does kernal_people return email, org, role?). This forces agents to guess or make extra discovery calls.
kernal_correct is dangerously under-constrained. The 'reset' action wipes all data with no confirmation step. Agents can accidentally destroy entire knowledge bases. Missing dry-run or confirmation pattern.
List/search tools (kernal_people, kernal_orgs, kernal_activities, kernal_actions) lack pagination documentation. No mention of how results are capped, whether there's a next_cursor, or what happens when results exceed limit. Baseline requires pagination for list tools.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | 2026-07-28+ | v2 |
Get a full briefing on a person or organization — timeline, network, topics, actions. Like a meeting-prep dossier.
Fix errors — update fields, delete entities, merge duplicates, or reset the entire database.
Create a relationship between two entities (people, organizations, or topics). Use for connections like "reports_to", "introduced", "knows", "works_at", "competes_with", etc.
List or search organizations in the knowledge base.
List or search contacts in the knowledge base.
Search the knowledge base for people, organizations, activities, topics, actions, or notes matching a query.
Store raw text and get instructions for extracting entities. Call this FIRST when the user shares meeting notes, call summaries, or any interaction. It saves the text and tells you which entities to extract and store using the add/link tools.
Error handling is absent. No guidance on what to do if a person/org doesn't exist, if a link target is missing, or if kernal_correct fails. Agents have no recovery path.
kernal_add_activity and kernal_link require entities to already exist ('must already exist in the knowledge base'). No guidance on how to handle missing participants or targets. Should suggest calling kernal_add_person first or provide a recovery path.