Drop your past contracts, playbooks, and house rules into Orchestra. We distil your terminology, risk tolerances, and drafting style — then every agent run inherits them. Reviews and drafts come back in your firm's voice, against your jurisdiction, on the LLM your InfoSec team approves.
Free for design partners · no credit card · we ship your firm's data isolation by Postgres RLS, not by promise.
We're built for firms that need their AI to reflect their own style, run on their own infrastructure, and produce an audit trail that survives a real review.
Drop your past NDAs, MSAs, employment agreements. We extract the firm's terminology, hard limits, and drafting style — and inject them into every agent's prompt. No prompt engineering, no copy-paste.
OpenRouter is the default — swap to Anthropic, OpenAI, Gemini, or Azure OpenAI by changing one row. Your API key is encrypted at rest and only decrypted at call time. InfoSec-friendly by design.
Postgres Row-Level Security on every tenant table. Every inference, tool call, pause, and resume lands in an append-only forensic log with a monotonic step number. Built for the next RBI / DPDPA review.
A single docker compose runs the whole stack. Exposes MCP so any agentic client (Claude Desktop, M365 Copilot) can plug in. Your data, your servers, your tokens — your choice.
Sign up, drop a zip, train. By tea time the agents speak in your firm's voice on your firm's terms.
Pick a slug, set an owner password, verify your email. Under 60 seconds.
Drag in a zip of NDAs, MSAs, playbooks. PDF, DOCX, TXT, MD — we parse, chunk, and index everything to pgvector.
Click Train. A map-reduce extraction distils your firm's house style and risk rules into a single override per agent.
Ask the agent to review a contract or draft a clause. Get a Word .docx back, in your firm's template, with your terminology.
NDA / MSA / employment / lease — flag clauses against your firm's playbook with green/amber/red classifications.
The agent fetches your golden template, fills in matter specifics, returns a tracked-change-ready DOCX.
Auto-flag non-Indian governing law, foreign arbitration seats, indemnity caps that exceed your limits.
Every inference, tool call, and HITL approval lands in an append-only log with monotonic step numbers.
We didn't add multi-tenancy and audit later. The platform won't boot without a tenant context. The audit log can't be skipped. The encryption can't be turned off.
Every row of every table is filtered by tenant_id at the database. An app bug cannot leak cross-firm.
Your LLM API key is Fernet-encrypted. Decrypted only at call time; never returned by the API.
Every inference, tool call, pause, and resume gets a monotonic step number per session. Replayable end-to-end.
Pre-tuned for INR, IST, Mumbai/Bengaluru arbitration, Indian governing law — not bolted on, baked in.
We're onboarding our first cohort of Indian firms now. We handle the setup. You bring the corpus and tell us what we should build next.