tinyhumansai (Python SDK)

Python SDK for TinyHumans Neocortex memory APIs (core memory, documents, mirrored routes, ingestion jobs).

Requirements

• Python 3.9+

Install

pip install tinyhumansai

For local development from this repo:

cd packages/sdk-python
uv sync --group dev

Get an API key

Export credentials before running examples:

export TINYHUMANS_TOKEN="your_api_key"

# Optional overrides:
export TINYHUMANS_MODEL_ID="neocortex-mk1"
export TINYHUMANS_BASE_URL="https://api.tinyhumans.ai"

Quick start

import time
import tinyhumansai as api

client = api.TinyHumansMemoryClient(token="YOUR_API_KEY")

# Insert (documents API) - document_id is required
namespace = "preferences"
document_id = f"pref-{int(time.time())}"
client.insert_document(
    title="User preference",
    content="User prefers dark mode",
    namespace=namespace,
    document_id=document_id,
)

# Recall context for an LLM prompt
ctx = client.recall_memory(
    namespace=namespace,
    prompt="What does the user prefer?",
    num_chunks=10,
)
print(ctx.context)

Notes:

• The legacy ingest_memory route maps key to the backend documentId.

Full route example

example.py exercises core + documents + mirrored routes + ingestion job polling + cleanup.

cd packages/sdk-python
python3 example.py

API reference

TinyHumansMemoryClient(token, model_id="neocortex-mk1", base_url=None)

Param	Type	Required	Description
token	str	✓	API key or JWT
model_id	str		Sent as X-Model-Id
base_url	str | None		Override API URL. If not set, uses TINYHUMANS_BASE_URL env or SDK default

client.ingest_memory(item={...})

Legacy ingest route. POST /v1/memory/insert

The SDK maps:

• key → backend documentId (and title)

Field	Type	Required	Description
key	str	✓	Unique key within the namespace
content	str	✓	Memory content
namespace	str	✓	Namespace
metadata	dict		Optional metadata
created_at	float | int		Unix timestamp (seconds)
updated_at	float | int		Unix timestamp (seconds)

Returns IngestMemoryResponse (ingested, updated, errors).

client.recall_memory(namespace, prompt, num_chunks=10, key=None, keys=None)

Query memory and return an LLM-friendly context string. POST /v1/memory/query

Field	Type	Required	Description
namespace	str	✓	Namespace
prompt	str	✓	Query prompt
num_chunks	int		Max chunks to retrieve
key	str | None		Optional single document id filter
keys	list[str] | None		Optional document id filters

Returns GetContextResponse (context, items, count).

client.delete_memory(namespace, delete_all=True)

Delete memory (namespace-scoped). POST /v1/memory/admin/delete

Field	Type	Required	Description
namespace	str	✓	Namespace
delete_all	bool	✓	Must be True to confirm deletion

Returns DeleteMemoryResponse (deleted).

client.insert_document(...)

Insert a single document. POST /v1/memory/documents

Field	Type	Required	Description
title	str	✓	Document title
content	str	✓	Document content
namespace	str	✓	Namespace
document_id	str	✓	Unique document ID
source_type	str | None		e.g. "doc"
metadata	dict | None		Optional metadata
priority	str | None		Optional priority
created_at	float | None		Unix timestamp (seconds)
updated_at	float | None		Unix timestamp (seconds)

Returns the backend data dict.

client.insert_documents_batch(items=[...])

Insert multiple documents in one call. POST /v1/memory/documents/batch

Field	Type	Required	Description
items	list[dict]	✓	Each item must include title, content, namespace, and documentId (or document_id)

Returns the backend data dict.

client.list_documents(namespace=None, limit=None, offset=None)

List documents. GET /v1/memory/documents

Field	Type	Description
namespace	str | None	Optional namespace
limit	int | None	Optional page size
offset	int | None	Optional page offset

Returns the backend data dict.

client.get_document(document_id, namespace=None)

Get document details. GET /v1/memory/documents/:documentId

Field	Type	Required	Description
document_id	str	✓	Document ID
namespace	str | None		Optional namespace

Returns the backend data dict.

client.delete_document(document_id, namespace)

Delete a document. DELETE /v1/memory/documents/:documentId

Field	Type	Required	Description
document_id	str	✓	Document ID
namespace	str	✓	Namespace

Returns the backend data dict.

client.query_memory_context(...)

Query memory context. POST /v1/memory/queries

Field	Type	Required	Description
query	str	✓	Query string
namespace	str | None		Optional namespace
include_references	bool | None		Include references
max_chunks	int | None		Optional chunk limit
document_ids	list[str] | None		Optional document filters
recall_only	bool | None		Recall-only mode
llm_query	str | None		Optional LLM query override

Returns the backend data dict.

client.chat_memory_context(messages=[...], temperature=None, max_tokens=None)

Chat with memory context. POST /v1/memory/conversations

Field	Type	Required	Description
messages	list[dict]	✓	Conversation messages [{role, content}]
temperature	float | None		Optional temperature
max_tokens	int | None		Optional token limit

Returns the backend data dict.

client.record_interactions(...)

Record interaction signals. POST /v1/memory/interactions

Field	Type	Required	Description
namespace	str	✓	Namespace
entity_names	list[str]	✓	Entity names
description	str | None		Optional description
interaction_level	str | None		Optional interaction level
interaction_levels	list[str] | None		Optional multiple interaction levels
timestamp	float | None		Optional Unix timestamp (seconds)

Returns the backend data dict.

client.get_ingestion_job(job_id)

Get ingestion job status. GET /v1/memory/ingestion/jobs/:jobId

client.wait_for_ingestion_job(job_id, timeout_seconds=30, poll_interval_seconds=1)

Poll an ingestion job until it reaches a terminal state.

Implemented methods

Core methods:

• ingest_memory
• ingest_memories
• recall_memory
• delete_memory
• recall_with_llm

Mirrored routes:

• chat_memory
• interact_memory
• recall_memory_master
• recall_memories
• query_memory_context
• chat_memory_context
• record_interactions
• recall_thoughts
• get_graph_snapshot

Documents:

• insert_document (requires document_id)
• insert_documents_batch (each item requires documentId / document_id)
• list_documents
• get_document
• delete_document

Ingestion jobs:

• get_ingestion_job
• wait_for_ingestion_job

Error handling

API errors raise TinyHumansError with message, status, and body.

import tinyhumansai as api
from tinyhumansai import TinyHumansError

client = api.TinyHumansMemoryClient(token="YOUR_API_KEY")
try:
    client.list_documents(namespace="ns", limit=1, offset=0)
except TinyHumansError as err:
    print(err.status, err, err.body)

Tests

cd packages/sdk-python
python3 scripts/test_routes.py