The memory OS
for AI agents.

A cognition engine that ranks, compresses and evolvesagent memory across any model, any framework, at any scale. Beyond storage. Beyond RAG.

Get started now

Memory infrastructure built for every agent you'll ever run

From solo developers to teams running hundreds of agents, Thrindex gives every agent the memory layer it needs to keep improving.

SDK

1
pip install thrindex
2
 
3
from thrindex import Thrindex
4
client = Thrindex(api_key="th_live_...")
5
 
6
# Store a memory
7
client.add(
8
    content="User prefers weekly email digests.",
9
    agent_id="support-agent",
10
    user_id="user-42",
11
)
12
# Search memories
13
results = client.search(
14
    query="how does the user want to be contacted",
15
    agent_id="support-agent",
16
    user_id="user-42",
17
)

API

1
import httpx
2
 
3
client = httpx.Client(
4
    headers={"Authorization": "Bearer th_live_..."}
5
)
6
 
7
# Store a memory
8
client.post("https://api.thrindex.com/v1/memories", json={
9
    "content": "User prefers weekly email digests.",
10
    "agent_id": "support-agent",
11
    "user_id": "user-42",
12
})
13
# Search memories
14
resp = client.post("https://api.thrindex.com/v1/memories/search", json={
15
    "query": "how does the user want to be contacted",
16
    "agent_id": "support-agent",
17
    "user_id": "user-42",
18
})
19
print(resp.json()["results"])

Intelligence before the query

Agents write and search. Cognition runs in the background - never while your agent waits.

Get started

Read in under 50ms

POST /v1/memories/search uses pre-ranked cache + multi-signal scoring. No LLM on the read path.

Read in under 50ms

POST /v1/memories/search uses pre-ranked cache + multi-signal scoring. No LLM on the read path.

Write in milliseconds

POST /v1/memories returns immediately. Dedup, importance, compression and conflict resolution run async.

Write in milliseconds

POST /v1/memories returns immediately. Dedup, importance, compression and conflict resolution run async.

Every agent memory, one dashboard

Browse, search and map what agents store - plus keys and analytics. Built for teams shipping AI in production.

Memory map

Your agent's entire memory, visible at once. See what's related, what's redundant and where knowledge is thin.

Memory map

Your agent's entire memory, visible at once. See what's related, what's redundant and where knowledge is thin.

Semantic search

Run the exact search your agent runs. See what it gets back, how fast and whether it came from cache.

Semantic search

Run the exact search your agent runs. See what it gets back, how fast and whether it came from cache.

Analytics

Latency, cache hits, ops volume and queue depth in one view. Memory you can actually monitor in production.

Analytics

Latency, cache hits, ops volume and queue depth in one view. Memory you can actually monitor in production.

Insights from our Blog

Discover useful articles about property protection, insurance basics, and real-world tips that help you get the most from your coverage.

View all

The agent remembered - it just remembered wrong

How an agent can recall a fact perfectly and still be wrong and what that costs the people who rely on it.

Technical Notes

Memory is not a search problem

A technical look at why retrieval is the easy half of agent memory and where the real engineering lives.

Technical Notes

The agent that remembered too much

Why keeping every memory slowly makes an agent worse and how forgetting becomes a feature.

Technical Notes

View all

GET STARTED

Creating memories

Give your agents persistent memory. Ranks what matters, compresses what doesn't, forgets what should be forgotten.

Get started

GET STARTED

Creating memories

Give your agents persistent memory. Ranks what matters, compresses what doesn't, forgets what should be forgotten.

Get started

GET STARTED

Creating memories

Give your agents persistent memory. Ranks what matters, compresses what doesn't, forgets what should be forgotten.

Get started