GEBRID PRE-TGE · BUILT ON BASE
The operating system for autonomous AI agents
GEBRID OS creates, deploys, coordinates and controls AI agents — through a Security, Guardian and Continuity control plane that enforces permissions and risk limits on-chain.
Why now
AI agents now move real money
AI agents market
$52.6B
projected by 2030
Up from $7.8B in 2025 — 46.3% a year (MarketsandMarkets). Every agent that touches capital needs limits it can’t talk its way past.
May 2026 · incident on Base
An AI agent was prompt-injected into sending ≈$150–200K in tokens
A gifted NFT escalated the agent’s permissions; a Morse-code message then got it to sign the transfer. No private keys were stolen (OECD AI Incidents Monitor).
The model’s reading of a prompt was enough to move funds. GEBRID’s rule: a model may propose, never authorize.
The problem
Owners have no single control layer
Agents already analyze data, write code, call tools and assemble financial operations without a human in the loop. Point products cover pieces. No system unifies the ten controls an owner needs.
Manage many agents centrally
Grant & revoke permissions
Isolated execution
Independent review of risky actions
State that survives migration
Tool & API access control
Spending limits
Verifiable action history
Identity & transferable ownership
Multi-agent coordination
The solution
One operating layer for AI agents
GEBRID OS creates, deploys, coordinates and controls autonomous agents — financial, analytical, enterprise and Web3 — inside limits that can be verified.
Governed autonomy
Agents that act on capital, apps and other agents — inside limits they can’t exceed.
Today: the on-chain security and permission core is built and tested locally. The full operating system ships in phases.
Design principle
AI proposes.
Code decides.
A language model never gets the right to approve a financial operation around mandatory checks. Permissions, budgets and risk limits are enforced by deterministic code at the smart-account level — not by prompts, and not by the model’s judgment.
How an action executes
Three mandatory gates before any action
- 01
Agent intent
The LLM proposes an action
- 02
Guardian & policy
Screens intent against owner rules
MANDATORY GATE - 03
Permission check
Signed, bounded, revocable
MANDATORY GATE - 04
Risk enforcement
RiskEngine decides on-chain
MANDATORY GATE - 05
Authorized execution
The smart account executes
- 06
Audit record
Verifiable action history
Fail any gate → the action never executes
Built and tested locally (Kernel v2 RC): the permission and risk gates on the on-chain path. Guardian AI screening is in research.
Architecture
Thirteen subsystems, one control plane
Status per subsystem as of 4 November 2026. Full integration of the operating system is not yet confirmed.
Guardian security layer
Three layers check every action
1 · Guardian AI
ResearchIndependent review of intent by two LLMs from different angles: manipulated instructions, swapped payment details, context that breaks owner limits.
2 · Guardian Policy
Partly builtFormal rules: limits, time windows, allowed recipients and functions, extra confirmation, emergency stop, revocation.
3 · Guardian Enforcement
Built · tested locallyContract permission checks and RiskEngine apply the final, mandatory decision on supported on-chain paths.
Guardian AI signals go to policy or a human; they never authorize a payment on their own.
Technical proof · Security Kernel v2 RC
189 contract tests passed. Zero failures.
189
contract tests passed
0
failures · 0 skipped
0/0
Slither High / Medium
3/3
SDK tests passed
Confirmed in local RC runs
- Authorized transfer executes
- Over-budget transfer is blocked
- Replayed operation is rejected
- Permission reuse is checked
- Balance controls hold
How it was tested
- Fuzz testing
- Stateful invariant testing
- Static analysis: Slither
- Local chain: Anvil, chain ID 31337
- Retained findings: 14 Low · 11 Informational · 1 Optimization
Source: internal RC report, 23 October 2026. An earlier enforcement core — 5 contracts and 127 tests (1 August 2026) — is available for code review under NDA. Nothing is deployed yet and no independent audit has been completed.
Who runs on GEBRID
Eight agent classes, one control layer
Each agent gets an identity, an owner, a budget and permissions it cannot exceed.
Trading agents
Market analysis and permitted trades
DeFi agents
Protocol actions within set permissions
Risk agents
Operation checks and risk scoring
Guardian agents
Independent control of other agents
Treasury agents
Approved corporate finance flows
Research agents
Collect, analyze and structure information
Enterprise agents
Workflow automation across systems
Developer agents
Coding, testing and maintenance
Target scenarios for the platform — not a list of shipped commercial products.
$GEBRID token
A liquidity-first, low-unlock design
Working model — Tokenomics v6 (23 October 2026), updated 4 November 2026: founders investment unlocks 10% at TGE. No token has been issued and the TGE is not scheduled.
Ticker
GEBRID (working)
Network
Base · EVM
Total supply
1,000,000,000
Proposed pair
GEBRID / USDT
TGE
Not scheduled
Contract
Pending final release
4.5%
of supply unlocks at TGE — 45,000,000 tokens
0% at TGE for team, seed, strategic/VC and treasury.
Planned market-making structure: the 30M stay in the treasury multisig and are loaned to a disclosed market maker under a written agreement — defined use, fixed term, unused tokens returned, no profit-sharing or guaranteed-return terms.
| Allocation | Share | Tokens | TGE | Cliff | Vesting |
|---|---|---|---|---|---|
| Founders investment | 5% | 50M | 10% | 12 mo | 48 mo |
| Pre-seed | 5% | 50M | 10% | 6 mo | 18 mo |
| Seed | 6% | 60M | 0% | 6 mo | 24 mo |
| Strategic / VC | 10% | 100M | 0% | 12 mo | 36 mo |
| Private | 5% | 50M | 5% | 6 mo | 12 mo |
| Public IDO | 2.5% | 25M | 10% | 1 mo | 12 mo |
| Team & advisors | 18% | 180M | 0% | 12 mo | 48 mo |
| Treasury / DAO | 12% | 120M | 0% | 12 mo | 48 mo |
| Protocol rewards | 12% | 120M | 0% | 6 mo | 60 mo |
| Ecosystem partners | 5% | 50M | 0% | 6 mo | 24 mo |
| Marketing / BD | 4% | 40M | 0% | 3 mo | 18 mo |
| Airdrop | 2.5% | 25M | 0% | 6 mo | 6 mo |
| Exchange liquidity & market making | 3% | 30M | 100% | — | — |
| Reserve | 2% | 20M | 0% | 24 mo | 24 mo |
| Unallocated reserve | 8% | 80M | 0% | not set | disabled |
| Total | 100% | 1,000M | 45M |
Marketing/BD (4%): a share unlocks at TGE for exchange launch campaigns; size set with the exchange’s requirements.
Months from TGE. The first linear release happens in the month after the cliff ends. All parameters are subject to final legal and technical review and to commitments already agreed with investors.
Why the network needs $GEBRID
Three planned sources of demand, each tied to a part of the security layer.
Stake to sign risk verdicts
Attesters lock GEBRID to publish RiskEngine attestations. A false or manipulated attestation is slashed.
Demand grows with the number of attesters and the value they secure.
Builds on ATTESTER_ROLE · plannedPay per security check
Each Guardian or RiskEngine check on paid tiers is billed in GEBRID; a share is burned or sent to the treasury.
Demand grows with agent activity: every checked action uses tokens.
PlannedSettle agent licences
Licences, rentals and transfers of agent identities (ERC-721 / ERC-6551) settle through GEBRID escrow.
Demand grows with marketplace volume.
PlannedAlso planned: governance of selected parameters and developer incentives.
None of these mechanics is live yet: the verified core runs without the token today, and staking, slashing, fee sharing and governance are planned. Nothing here is an offer to sell or a solicitation to buy tokens, and no price or return is promised.
Business model
Seven revenue lines on one platform
GEBRID OS SaaS
Subscription to the agent operating environment
Enterprise Agent Infrastructure
Integration with corporate systems and finance tools
Guardian Security Services
Independent control, policy and action checks
Developer SDK & API
Licensing of infrastructure modules
Agent Marketplace
Fees on licensing, transfer and use of agents
Risk-as-a-Service
Risk enforcement for external agent platforms
Infrastructure Partnerships
Joint builds with exchanges, wallets and AI providers
Pre-revenue
These are planned models. No ARR, MRR or paying-user revenue is claimed.
Intellectual property
Ten directions under patent work
Patent families
GEBRID-001
GEBRID-002
Filing status, ownership and USPTO documents are shared in due diligence once official confirmations are reconciled.
Applications are not presented as granted patents.
Roadmap
Seven phases from kernel to ecosystem
- PHASE 1
Foundation
Identity, permissions, RiskEngine
Built · local tests - PHASE 2
Public Security Preview
Base Sepolia deployment and scenarios
Next - PHASE 3
OS Integration
Kernel, orchestrator, runtime
In development - PHASE 4
Guardian & Continuity
Guardian AI, memory, MCP security
Prototypes - PHASE 5
Commercial Pilots
Wallets, AI platforms, enterprise
Qualifying partners - PHASE 6
Token & TGE
Final contract, review, market making, exchanges
Preparation - PHASE 7
Ecosystem
Marketplace, licensing, SDK
Product roadmap
Phase dates are set once technical readiness and resources are confirmed.