Claiv

Software for AI that has to account for itself.

claiv.io · How this got here · Docs

Three products, built in that order, each the answer to a problem the last one exposed. The full account is at claiv.io/about.

An AI organisation you can govern. Describe an outcome and Studio designs a team of specialists around it — their knowledge, their connections and their limits set out before any of it exists. You approve the design. It builds the specialists and proves them against realistic cases. Then it runs, and every decision, authority and verified effect is on the record.

• Authority is explicit per specialist
• Approval is bound to the exact action, not to a category
• Effects are verified against the system itself, not assumed from an approval
• Credentials are handles, never values
• Runs on your own computer, and can run models locally

Windows, in paid early access. Request access → The download and every release note live on your account page.

A persistent memory API. Your AI forgets everything between sessions; context windows only move the ceiling. Memory gives it continuity:

• Persistent user memory across conversations
• Document understanding, not just chunk retrieval
• Temporal awareness — what happened, and when
• Context injection ready for an LLM prompt
• Controlled forgetting, for when someone asks to be deleted

POST /v6/ingest      →  store memory events
POST /v6/recall      →  retrieve context
POST /v6/documents   →  upload documents
POST /v6/forget      →  delete memory

Python

pip install claiv-memory

from claiv import ClaivClient

client = ClaivClient(api_key="your_key")

client.ingest({
    "user_id": "user_123",
    "conversation_id": "conv_abc",
    "type": "message",
    "role": "user",
    "content": "I'm building an AI agent for compliance review.",
})

context = client.recall({
    "user_id": "user_123",
    "conversation_id": "conv_abc",
    "query": "What is the user building?",
})

print(context["llm_context"]["text"])   # inject into your system prompt

JavaScript / TypeScript

npm install @claiv/memory

import { ClaivClient } from '@claiv/memory';

const client = new ClaivClient({ apiKey: 'your_key' });
await client.ingest({ user_id: 'user_123', conversation_id: 'conv_abc',
  type: 'message', role: 'user', content: "I'm building an AI agent." });
const context = await client.recall({ user_id: 'user_123',
  conversation_id: 'conv_abc', query: 'What is the user building?' });

Official LoCoMo result across all ten dialogues — the standard benchmark for long-conversation memory:

• Ingest is asynchronous — memory may not appear instantly
• Always include user_id and conversation_id
• Inject llm_context.text into your system prompt, not the raw events
• API key authentication with tenant isolation; deletion is scoped via /v6/forget

Version 6.3, in production at api.claiv.io. Available on request rather than self-service sign-up — read the docs, then write to support@claiv.io and we will set up a project and issue a key.

Deterministic structured output, with no model. Every system that asks an LLM for JSON eventually writes half a parser to cope with what comes back: fenced blocks, chatty preambles, Python True, truncation — and search models that return prose with citations and no JSON at all.

Six tiers run cheapest-first and stop at the first schema-valid result. Tiers 0–4 are pure functions, so the same input and schema give byte-identical output every time, offline. It never invents: anything absent from the source is null and listed in unresolved, and every field reports how it was resolved — exact, repaired, coerced, or read from markdown.

npm install @claiv/parser

MIT, zero-dependency · npm