Agentic Systematic Literature Reviews
Thoth
An agent that drafts evidence-grounded literature reviews — and verifies every cited claim against the source paper before you read the draft.
Named for Thoth, ancient Egypt’s ibis-headed god of writing, wisdom, and scribes — the divine patron of the work this tool automates.
No sign-up. We mint a 24-hour guest account preloaded with the ReAct paper review — draft, scores, and the full cite_check audit.
What makes Thoth different
Four design choices that separate Thoth from generic “AI research” tools.
cite_check post-pass
Every cited claim is verified against the source paper before the draft reaches you. Hallucinated citations are flagged, not hidden.
Authenticated MCP server
OAuth 2.1 + PKCE + Dynamic Client Registration via Clerk, with SHA-256 audit logs. Listed in the official MCP Registry.
Public eval dashboard
Every commit can run the agent against a versioned golden set; results render at /evals. Regressions are visible, not buried.
6 LLM providers, $0 default
Swap providers with one env var. Mistral free tier is default. Local eval runs ride a Claude Max subscription via the agent SDK.
Verified engineering proofs
io.github.ahmedEid1/thoth199 tests · tsc · lint · all greenSHA-256 audit log on every MCP call · no raw input stored$0 / month deploy on free tiersSpec-driven build: specs → plans → TDD subagents → reviewed → shipped. v0.7.0-m5 live. M6 (30-question eval set + public launch) next.