On June 15, 2025, at a closed-door data summit in Silicon Valley that was not live-streamed, Kite CEO Chi Zhang presented a chilling architectural diagram:
“Today, 90% of AI agents are just ghost addresses on-chain with no 'past'.”
There is no traceable data lineage, which in the traditional corporate world means — this data cannot enter reports;
There is no traceable behavioral lineage, which in the Web3 machine economy means — this agent is not fit to manage money.
As a former core architect at Databricks, Chi is very clear: lineage is order.
Kite's SPACE framework essentially transplants Databricks' entire data governance philosophy into the on-chain identity and permission system—providing each 0x address with a verifiable "birth certificate" and "growth record".
PayPal Ventures' willingness to lead a $33 million investment isn't based on a new public blockchain, but rather on this metaphor:
Kite is not creating another AI meme, but rather establishing a "digital household registration system" for AI agents.
From Data Lineage to Proxy Lineage: SPACE's On-Chain Implementation
In the world of Databricks, Lineage solves a very simple problem:
Which tables were used to calculate this metric?
Which ETL task corrupted the model?
Who should be blamed for this misjudgment?
Kite replaced the same set of questions with an Agent context:
Is this 0x address really the company's own business agent?
Is the order it placed today a continuation of its established strategy, or was it swayed by the Prompt Injection?
Is a huge loss a model problem, a data problem, or a "hallucination accident" in a single session?
What the SPACE framework does is transform these question marks into "lineage paths" that can be reconstructed on the chain:
Who created whom, who authorized whom, and in what context, which decision was made?
For AI Agents, this is the structured form of the digital soul.
BIP-32 Family Tree: Who is responsible for the decisions made by the AI agent?
Kite did not follow the old path of "one address for everything" at the account level, but instead directly used a modified version of the BIP-32 layered deterministic wallet.
The structure is tree-like:
Root node: User
It possesses the highest authority and is the "master" of the entire tree.
Child node: Agent (proxy)
For different business areas: transaction agency, risk control agency, procurement agency... each operates independently.
Leaf node: Session
A specific task, a specific Prompt context
Each layer has its own key, and each layer also has clearly defined boundaries of responsibility.
This design addresses a very specific and also very harsh problem:
When a financial agent loses money due to a model illusion, where should they cut their losses?
Does this mean canceling the quota for this session?
Is this the version of the policy that disables this agent?
Or should we freeze the root user permissions for the entire tree?
With the lineage tree, the system can accurately answer the question of the path of "responsibility transmission"—
From Prompt, to model version, to specific agent, to the final asset account.
For regulators, this is exactly what they want to see: algorithm accountability.
For businesses, this is the prerequisite for truly daring to entrust "real money" to AI.
Between Privacy and Trust: Folding Trust Costs with Zero Knowledge
Simply applying Databricks-style Lineage to blockchain will lead to a huge pitfall:
Traditional Lineage = The more transparent, the better
The business world = too transparent = trade secrets exposed
Kite's biggest challenge wasn't how to remember everything, but rather:
How can we selectively forget things and show them to outsiders, while still being able to remember them?
Their solution involves introducing zero-knowledge proofs (ZKP):
For the partners:
The agent can prove – "I belong to a Fortune 500 company that has passed KYC."
However, it will not reveal who the company is, nor will it reveal its complete internal structure.
For oneself:
Lineage details are preserved in a verifiable encrypted structure.
In the event of a real incident, the company can unlock internal views to achieve "precise accountability and external cover-up."
The result is:
Trust-based interaction stems from "lawyers + contracts + multiple rounds of due diligence".
It has become "ZK proof + one on-chain verification"
This isn't about sentimentality; it's about real conversion rates.
After adopting this identity architecture, the success rate of cross-organizational collaboration for enterprise-level AI agents skyrocketed from approximately 15% to around 85%.
Because neither side needs to use a bunch of NDA heaps of trust anymore, but instead uses mathematics.
Let tokens carry "credit scores": the core value of KITE
On this structural foundation, the KITE token is no longer just a "Gas + payment tool," but has been given a more important role: a carrier of identity and credibility.
The logic is roughly as follows:
Every verified agent's actions on the network are recorded:
Timely settlement, no malicious operations, and adherence to credit limits → Positive points awarded.
Frequent chargebacks, abuse of authority, triggering risk control measures → negative score recorded.
These behaviors accumulate on the blockchain to form a "reputation curve".
The reputation curve then flows back to the upper-level user accounts, transforming into:
Higher credit limit
Lower settlement friction
Priority task allocation rights
in other words:
KITE is more than just a tool for you to "pay".
It's still a "pledge certificate" that the system uses to score you and increase your credit limit.
For long-term operators, this mechanism rewards consistent, law-abiding, and stable contributions;
For short-term speculators who rely on hit-and-run tactics, this mechanism is tantamount to slamming a cold, dismissive statement at the door:
"You can come and masturbate, but once you're done, it'll be hard for you to come back."
This is a standard enterprise service gene:
Rewards for "growing big and lasting"
Punish those who "take all the money and run away"
From Databricks to Kite: Writing Order into the Machine Economy
If you extend the timeline, you'll find that Chi Zhang has actually been doing the same thing all along:
At Databricks, he provides order to chaotic data flows.
In Kite, he provided order for autonomous proxy networks.
But this time,
The data is no longer just Parquet files passively lying in the warehouse.
Instead, it's an AI agent that places orders, negotiates prices, and calls APIs on its own.
And order is no longer the process specifications written in an internal corporate wiki.
Instead, they are mathematical constraints written into smart contracts, wallet topologies, and ZK proofs.
On testnet v2, the identity tree of 17,600 users is slowly growing taller:
Some are Side Hustle proxies uploaded by individual developers.
Some are Settlement Agents and Risk Agents tested by large enterprises.
Each node is a "digital citizen" that is building trust, accumulating assets, and expanding the boundaries of its authority.
For Binance CreatorPad readers, there is only one key signal here:
Kite is not a "chain AI concept stock" propped up by a story and TVL.
Instead, it's a complete system validated by Silicon Valley's data engineering framework.View of order，
Projection in the Web3 machine economy.
As the next wave of AI agent narratives shifts from "fun" to "truly managing money,"
Whoever possesses the three elements of lineage, behavioral lineage, and economic incentives...
Whoever is qualified is the "root system" of this track.
I'm like someone trying to find a lost sword by marking the boat, observing the power landscape of the machine economy over the next few years on each branching address family tree. @KITE AI $KITE #KITE
