Loading ToolBench
OneBrain persistent memory integration for OpenClaw — auto-capture, recall, and search across sessions
OneBrain OpenClaw provides 5 memory-focused tools with clear verb-noun naming (recall_context, search_memories, save_memory, get_entities, get_context). All tools have descriptions (avg ~120 chars, within baseline 34-392 range) and input schemas with type definitions. However, output schemas are completely undocumented, no structured response format is specified for any tool. Error handling is minimal (generic try-catch blocks returning error strings). Parameter descriptions are present but lack actionable constraints (e.g., no guidance on confidence score interpretation, no enum validation for scope values despite enum declaration). The save_memory tool accepts a 'confidence' parameter but provides no guidance on what confidence means or how it affects recall ranking.
Retrieve formatted context about the user — their preferences, goals, recent decisions, and key entities. Use this to understand the user's background and current situation.
List known entities — people, projects, tools, organizations. Use this to understand who and what the user works with.
Get relevant memories for the current context. Use this to recall past decisions, preferences, facts, and experiences related to a topic.
Store new information as a persistent memory. Use this to save facts, preferences, decisions, goals, experiences, or skills for future recall.
Search stored memories with filters. Use this to find specific memories by content and type.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract structured data from responses. All 5 tools return formatted text strings with no declared field structure.
save_memory confidence parameter lacks actionable guidance. Description says '0.0 to 1.0' but does not explain what confidence means (user certainty? data quality? recall weight?) or how it affects future recall ranking.
get_context scope parameter declares enum ['brief', 'assistant', 'project', 'deep'] but description does not explain what each scope returns or when to use each. LLM must guess.
Error handling is generic. All tools catch errors and return 'Error: <message>' strings. No recovery guidance, no error classification (retryable vs fatal), no actionable next steps for LLM.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2026-07-28+ | v2 |
No pagination support. search_memories and recall_context accept limit but do not return total_count, next_cursor, or offset. Large result sets will be truncated silently.