Loading ToolBench
The open-source platform for running an Agentic Company. A CLI tool and framework for scaffolding AI agent teams, managing orchestration, knowledge bases, and multi-channel deployment.
Tellet provides 13 tools with complete JSON schemas and descriptions for all tools and parameters. Naming follows verb_noun conventions (list_, get_, update_, delete_, search_, add_, run_, manage_, install_). However, descriptions are generic and lack LLM-optimization guidance on WHEN to use each tool vs. alternatives. Parameter descriptions are present but minimal (avg 30-50 chars vs. baseline 72 chars). No output schemas documented. Error handling and recovery guidance absent. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk classifications. Missing dependency hints and multi-step guidance. Schemas are well-formed but lack constraints (enums, patterns, min/max bounds) that would prevent hallucinated values.
Add a document to the company knowledge base. Agents use this to answer customer questions accurately.
Delete a document from the knowledge base by ID.
Get recent conversations with message counts and agent info.
Get company statistics: total conversations, messages, active agents, and estimated cost.
Install a tool from the marketplace and assign it to agents. Updates tellet.json. The owner will need to set the required API key in .env.
List all AI agents in the company with their name, role, status, and model.
No output schemas documented. LLMs cannot plan downstream tool calls or extract required fields (e.g., agent_id from list_agents for use in update_agent_prompt). Baseline: 100% of A+ tools document return types.
Missing tool annotations despite clear risk classifications. delete_knowledge (DESTRUCTIVE), update_agent_prompt (WRITE), run_agent_task (WRITE) lack destructiveHint/idempotentHint. Agents cannot reason about retry safety or confirmation requirements.
Descriptions lack LLM-optimization guidance. Generic descriptions (avg 60-65 chars) do not explain WHEN to use each tool vs. alternatives. E.g., 'Get recent conversations' does not clarify: does this return unread only? Sorted by date? Does it include message content or just metadata?
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
List all tools available in the tellet marketplace that can be installed.
List all documents in the knowledge base.
Enable or disable scheduled tasks for an agent. Updates tellet.json.
Run a specific agent with a task immediately. Use this when the owner asks to have an agent do something right now.
Search the knowledge base to check what information agents have access to.
Update an agent's system prompt. Use this when the owner wants to change how an agent behaves.
Update the website content. Can update tagline, subtitle, features, FAQ, or CTA. Only include fields that need to change.
No input validation constraints. Parameters lack enums, patterns, min/max bounds. E.g., 'limit' in get_recent_conversations has no bounds (could be 0, -1, 999999). 'category' in add_knowledge lists examples in description ('policy, product, faq, or general') instead of enum constraint.
No error handling or recovery guidance. Tools do not document what errors can occur, whether they are retryable, or what the LLM should do next. E.g., update_agent_prompt does not explain: what if agent_id is invalid? What if system_prompt exceeds length limits?