v0.1.1 — Recursive Reasoning

The Whole
in Parts.

Recursive language model reasoning for documents. Iterate with search, code, and think tools. Converge on answers with provenance.

Get Started View on GitHub
א

The Process

Load. Explore. Cite. Converge.

Load → Search → Cite → Evaluate → Finalize
↻ The Recursive Loop
MCP Server

Provenance
Tracking.

Every search, every citation, every piece of evidence is logged automatically. When you finalize, you get a complete audit trail.

  • <> exec_python — Run code against context
  • ⌕ search_context — Regex search with evidence logging
  • ✓ cite() — Tag findings with notes
  • ◉ evaluate_progress — Track convergence

EVIDENCE TRAIL

⌕ [search] pattern: "liability" lines 42-58
✓ [manual] note: "key exclusion clause"
⌕ [search] pattern: "indemnif" lines 112-130
✓ [manual] note: "carve-out for negligence"

CONFIDENCE: HIGH (0.92)

→ Ready to finalize with 4 citations

The Recursive
Method.

Aleph is inspired by research on Recursive Language Models. Rather than solving context limits at the architecture level, let models partition context and make recursive calls to themselves.

"Each iteration lets the model refine its search based on what it learned, rather than betting everything on one attention pass."
"What my eyes beheld was simultaneous, but what I shall now write down will be successive, because language is successive."

— Jorge Luis Borges, "The Aleph" (1945)