Observable Agency
for AI Systems
Thoughtbox is an intention ledger. When an agent causes a failure, you get the black box recording — what it was reasoning, what it considered, and why it decided what it did.
CONNECT IN SECONDS
{
"mcpServers": {
"thoughtbox": {
"type": "http",
"url": "https://mcp.kastalienresearch.ai/mcp?key=tbx_YOUR_API_KEY"
}
}
}THE AUDIT TRAIL
YOUR AGENTS MISS
AGENTS ARE EPHEMERAL. THOUGHTBOX CAPTURES THE REASONING THAT DISAPPEARS WHEN THE SESSION ENDS.
Replayable reasoning traces
Every session is a numbered, timestamped chain of thoughts. Reconstruct exactly what the agent was reasoning about when it took any action.
Branch visualization
See where the agent explored alternatives, forked into parallel paths, and converged on a decision. Topology is visible, not hidden.
Code Mode
Two MCP tools replace dozens. Agents write JavaScript against the tb SDK to search, record, and analyze — no context window bloat.
Knowledge graph
Entities, relations, and observations accumulate across sessions. Prior experience surfaces automatically to new agents.
KNOW WHY THEY DECIDED WHAT THEY DID
FREE THROUGH MAY 10. FULL ACCESS, NO CREDIT CARD. START RECORDING AGENT REASONING TODAY.
JOIN THE FOUNDING BETA