Operational automation and on chain agents.
Automated systems that monitor, execute, and manage on chain operations. Keepers, bots, alerting pipelines, and AI driven decision logic that operates transparently. The agent acts, and you can always see what it did and why.
What we build.
Every automation system starts with a clear operational envelope. These are the agents, keepers, and infrastructure we deploy most often, and the problems where automated execution delivers the most value.
On Chain Keeper Networks
Automated transaction execution triggered by on chain conditions. Time based or condition based triggers that fire reliably, with redundancy and gas management built into the infrastructure from the start.
- Chainlink Automation integration for condition based upkeep
- Gelato relay networks for gasless and sponsored execution
- Custom keeper infrastructure with failover and priority queues
- Time based cron triggers for scheduled on chain operations
Liquidity Management Agents
Automated rebalancing of liquidity pool positions across protocols and venues. Concentrated liquidity range adjustment, yield optimization, and position management that responds to market conditions without waiting for a human to wake up.
- Concentrated liquidity range rebalancing for Uniswap V3 style pools
- Multi venue position management across lending and AMM protocols
- Yield optimization agents that route capital to highest risk adjusted returns
- Automated compounding and reward harvesting across DeFi positions
Risk Monitoring and Liquidation
Collateral health tracking, automated liquidation triggers, and position unwinding logic designed to protect protocol solvency. These systems monitor health factors continuously and execute precisely when thresholds are crossed, with cascade prevention built into every trigger.
- Collateral ratio monitoring with configurable health factor thresholds
- Automated liquidation bots with partial liquidation support
- Cascade prevention logic that limits systemic liquidation chains
- Position unwinding with slippage protection and price impact limits
Treasury Automation
Scheduled token distributions, treasury diversification execution, payroll automation, and grant disbursement. These agents handle the recurring financial operations that DAOs and protocols need to run consistently, with every transaction logged on chain for full auditability.
- Scheduled contributor payments and vesting distributions
- Treasury diversification execution across stablecoins and yield bearing assets
- Grant disbursement automation tied to milestone verification
- Gas wallet top up agents that maintain operational reserves
Fraud Detection Systems
Real time transaction pattern analysis that identifies suspicious behavior as it happens. Velocity checks, anomaly detection, and automated flagging that catches coordinated abuse, wash trading, and withdrawal manipulation before losses compound.
- Transaction velocity analysis with adaptive baseline thresholds
- Pattern matching for coordinated wallet activity and sybil behavior
- Automated flagging and temporary blocking of suspicious addresses
- Withdrawal pattern monitoring for gaming and exchange operators
Arbitrage and MEV Infrastructure
Cross venue price monitoring and execution strategies that capture legitimate arbitrage opportunities. MEV protection mechanisms that shield users from sandwich attacks and front running, integrated with private transaction pools and flashbot relays.
- Cross DEX price monitoring with sub second latency
- Atomic arbitrage execution through flash loans and multicall batching
- MEV protection for user transactions via private mempools
- Flashbots and MEV Share integration for fair ordering
Incident Response Pipelines
Automated alerting, escalation chains, and emergency response mechanisms that activate the moment something goes wrong. Circuit breakers that pause contracts, emergency withdrawal triggers, and structured post incident reporting that captures exactly what happened and when.
- Circuit breaker contracts with automated pause and rate limiting
- Escalation chains routed through PagerDuty, Slack, and custom webhooks
- Emergency pause triggers with timelocked recovery procedures
- Structured post incident timelines with on chain event correlation
Problems this solves.
These scenarios reflect the types of automation engagements that deliver the most value. The operational contexts vary, but the pattern is consistent. Define the envelope, build the agent, log everything.
Automated liquidity rebalancing for a DeFi protocol
A DeFi protocol running concentrated liquidity pools across multiple venues needs positions rebalanced as price moves outside profitable ranges. This requires agents that monitor tick ranges continuously, evaluate gas costs against expected returns, and execute rebalancing transactions only when the math justifies the move. The agent manages dozens of positions across venues, compounding fees and adjusting ranges without human intervention while logging every decision on chain.
Automated liquidation with cascade prevention for a lending platform
A lending platform needs to liquidate undercollateralized positions quickly enough to protect protocol solvency, but carefully enough to avoid triggering cascading liquidations that destabilize the entire market. This requires health factor monitoring across thousands of positions, partial liquidation logic that minimizes market impact, and circuit breakers that pause liquidations if systemic risk thresholds are approached. The bot simulates every liquidation before execution to verify acceptable slippage.
Fraud detection agents for a gaming operator
A gaming operator running on chain casino games needs to detect coordinated fraud patterns in real time. Players attempting bonus abuse, multi account withdrawal manipulation, and collusion patterns need to be flagged before losses compound. This requires agents that monitor withdrawal velocity, cross reference wallet clustering, track session timing anomalies, and automatically flag or temporarily restrict suspicious accounts while generating evidence packages for compliance review.
Treasury distribution automation for a DAO
A DAO with dozens of active contributors needs automated monthly payments, grant milestone disbursements, and treasury diversification execution. Manual multisig ceremonies for routine payments create operational bottlenecks and delay contributor compensation. This requires keeper agents that execute approved payment schedules on chain, verify milestone conditions before releasing grant tranches, and manage treasury rebalancing within governance approved parameters.
Collateral monitoring and rebalancing for a stablecoin issuer
A stablecoin issuer needs continuous monitoring of collateral ratios across multiple backing assets, with automated alerts when ratios approach critical thresholds and automated rebalancing when governance approved parameters allow. This requires agents that track asset prices across oracles, calculate real time collateralization ratios, trigger alerts through multiple channels when thresholds are crossed, and execute rebalancing transactions within tightly scoped permissions.
How on chain agents work.
An on chain agent is not a script running in a cron job. It is a carefully scoped system with a defined trigger, evaluation logic, execution path, and recovery mechanism. Every component is designed for transparency, redundancy, and bounded failure.
Trigger, Evaluate, Execute
Every agent follows the same fundamental cycle. A trigger fires when an on chain condition is met or a time interval elapses. The evaluation layer determines whether the action should proceed based on current state, gas costs, and risk parameters. Only then does the execution layer submit a transaction. This separation prevents the common failure mode where an agent blindly executes regardless of context.
The evaluation step is where most of the complexity lives. A rebalancing agent does not just check if a position is out of range. It checks gas costs, expected slippage, competing transactions in the mempool, and whether the rebalance will actually improve returns after fees. Only when all conditions pass does the agent submit.
Keeper Network Design
Production keeper infrastructure requires redundancy. A single keeper node is a single point of failure. The system needs multiple keeper instances with coordination logic that prevents duplicate execution while ensuring at least one instance picks up every eligible task. This means consensus on task assignment, nonce management across keepers, and fallback promotion when a primary keeper goes silent.
Gas management is its own discipline. Keepers need gas funding strategies, dynamic gas price bidding that adapts to network congestion, and priority fee logic that balances execution speed against cost. For time sensitive operations like liquidations, the keeper may need to bid aggressively. For routine distributions, it waits for cheap gas windows.
Off Chain Computation with On Chain Execution
The hybrid pattern is the backbone of sophisticated automation. Complex computation happens off chain where resources are cheap and flexible. The results are submitted on chain where execution is verifiable and immutable. This pattern applies to everything from yield optimization to fraud scoring. The AI or algorithm runs off chain, produces a recommendation, and the on chain contract validates and executes within its defined parameters.
The critical design constraint is that the on chain contract never trusts the off chain computation blindly. It verifies that the proposed action falls within its allowed envelope. If the off chain layer recommends a rebalance that exceeds the maximum position size, the contract rejects it. The chain is the final authority.
Agent State Management and Recovery
Agents maintain state, and state can become corrupted. A keeper that crashes mid transaction may leave a pending nonce that blocks subsequent executions. A monitoring agent that loses its checkpoint may re process events or skip them entirely. Robust state management means persisting checkpoints, implementing idempotent execution, and building recovery procedures that restore the agent to a known good state without manual intervention.
Every agent has a heartbeat. If the heartbeat stops, the monitoring layer detects the silence and triggers an alert. Dead agents are worse than no agents, because teams stop watching for the things the agent was supposed to handle.
Transaction Simulation Before Execution
Every transaction is simulated against the current chain state before it is submitted. Simulation catches reverts, unexpected gas costs, slippage exceeding thresholds, and state changes that would make the transaction unprofitable or harmful. This is not optional. A liquidation bot that submits without simulation risks reverting at the worst possible time and losing the gas fee while the position it should have liquidated deteriorates further.
Simulation also serves as a final validation layer. If the simulation produces a result that differs from what the evaluation layer predicted, the agent halts and alerts. Discrepancies between prediction and simulation are a signal that something about the system state has changed unexpectedly.
Gas Optimization and Batching
When an agent manages many positions or executes many small transactions, batching becomes essential. Multicall contracts aggregate multiple operations into a single transaction, reducing per operation gas costs significantly. Batching also enables atomic execution, where either all operations in the batch succeed or none do, which is critical for operations like multi position rebalancing where partial execution could leave the system in an inconsistent state.
Gas timing strategies further reduce costs. Non urgent operations queue until gas prices drop below a configured threshold. The agent maintains a priority system where time sensitive operations execute immediately at market gas rates while routine maintenance operations wait for optimal conditions.
AI suggests, the chain executes.
AI and machine learning are powerful tools for generating parameters, strategies, and optimization logic. The key principle is that AI operates in the advisory layer while the blockchain remains the execution and audit layer. Every decision is logged. Every execution is verifiable.
Strategy Generation and Optimization
AI models analyze historical on chain data, liquidity patterns, and market microstructure to generate trading parameters and optimization strategies. A yield optimization agent might use ML to predict which lending pools will offer the best risk adjusted returns over the next period, while a rebalancing agent uses reinforcement learning to optimize its trigger thresholds based on backtested performance.
The AI generates recommendations. The on chain contract validates that every recommendation falls within governance approved bounds before executing. The model can suggest moving capital to a different pool, but it cannot override the maximum position size or approved protocol list hardcoded in the contract.
Backtesting and Continuous Learning
Every AI driven strategy runs through a backtesting framework before deployment. Backtests replay historical chain state and simulate how the agent would have performed under real conditions, including gas costs, slippage, and competing transactions. Strategies that pass backtesting enter a shadow mode where they run alongside live operations without executing, allowing comparison between AI recommendations and actual outcomes.
Once live, the system tracks performance metrics continuously and adjusts parameters through a controlled feedback loop. Parameter updates go through the same governance approval process as initial deployment. The AI learns and adapts, but never autonomously expands its own authority.
Transparency and Audit Trails
Every AI decision is logged with full context. When an agent rebalances a position, the log includes the model's input data, the parameters it generated, the evaluation result, and the execution outcome. This creates a complete audit trail that lets anyone reconstruct why a particular action was taken at a particular time.
- Decision logs stored on chain via event emissions and IPFS anchoring
- Model version tracking tied to each execution for reproducibility
- Performance attribution showing AI contribution versus baseline
- Anomaly flags when AI recommendations deviate significantly from historical patterns
Guardrails and Human Override
AI driven agents operate within hard constraints that no model output can override. Maximum transaction sizes, approved protocol lists, position concentration limits, and loss thresholds are enforced at the contract level. If an AI model produces an output that would violate any guardrail, the contract rejects the transaction and alerts the operations team.
- Hard coded position limits and loss thresholds in smart contracts
- Human override capabilities for pausing, adjusting, or terminating agents
- Graduated autonomy model where agents earn expanded envelopes through demonstrated performance
- Kill switch accessible to authorized multisig signers at any time
Visibility into every operation.
Automation without monitoring is a liability. The systems we build include comprehensive observability, alerting, and incident response infrastructure so that when something unexpected happens, the right people know immediately and have the context to respond.
System Health Dashboards
Every deployed agent gets a real time dashboard showing operational status, execution history, gas consumption, success and failure rates, and performance metrics. Dashboards are built in Grafana with Prometheus backends, or as custom interfaces when the operational team needs domain specific views. The dashboard is not decoration. It is the primary interface the operations team uses to understand whether the system is behaving as expected.
- Agent heartbeat and uptime tracking
- Transaction success rates with failure categorization
- Gas consumption trends and cost attribution
- Position and balance summaries across managed protocols
On Chain Event Monitoring
Dedicated watchers subscribe to on chain events that matter for the system. Contract state changes, large transfers, governance proposals, oracle price deviations, and unusual transaction patterns all generate structured events that feed into the monitoring pipeline. Event processing handles chain reorganizations, missed blocks, and indexing delays so that monitoring stays accurate even when the underlying chain is unstable.
- Event indexing with reorg handling and delivery guarantees
- Configurable filters for event severity and relevance
- Correlation of on chain events with off chain agent actions
- Historical event replay for investigation and debugging
Alert Configuration and Routing
Alerts are configured with clear severity levels and routed to the appropriate channels. Critical alerts go to PagerDuty and trigger on call rotations. Warning level alerts go to dedicated Slack channels. Informational alerts feed into email digests. Custom webhooks integrate with any existing incident management tooling the team already uses. Every alert includes actionable context, not just a threshold breach notification.
- Severity tiered routing to PagerDuty, Slack, email, and webhooks
- Alert deduplication and grouping to prevent notification fatigue
- Contextual alert payloads with relevant transaction data and links
- Configurable quiet hours with escalation bypass for critical alerts
Runbook Automation and Post Incident Analysis
Common incident responses are codified into automated runbooks. When a known alert pattern fires, the runbook executes the first response steps automatically, gathering diagnostic data, pausing affected systems if needed, and preparing the context that a human responder needs to take over. After every incident, structured post incident reports capture the timeline, root cause, and follow up actions with references to the on chain events and agent logs that tell the full story.
- Automated diagnostic data collection on alert trigger
- Predefined escalation procedures with role based notification
- Post incident report templates with on chain event correlation
- Continuous improvement tracking for recurring incident patterns
Tools we use daily.
We choose automation infrastructure based on reliability requirements, cost constraints, and the specific operational context. These are the platforms, frameworks, and services we deploy across our automation engagements.
Keeper and Relay Networks
For production keeper infrastructure, we integrate with established networks for reliability and use custom keepers when the operational requirements demand more control. Chainlink Automation provides battle tested condition based execution. Gelato handles gasless relaying and cross chain triggers. OpenZeppelin Defender provides a managed platform for monitoring and automated transactions.
Monitoring and Security
On chain monitoring combines dedicated threat detection networks with transaction simulation platforms. Forta provides decentralized threat detection with community maintained detection bots. Tenderly offers transaction simulation, alerting, and debugging tools. Together they form the foundation of the observability layer that keeps automated systems transparent and safe.
Agent Runtime and Orchestration
Agents run as containerized services orchestrated through Kubernetes or serverless functions depending on execution frequency and latency requirements. Python handles ML model inference and data analysis pipelines. Node.js and TypeScript power event driven agent loops and API integrations. Both connect to on chain execution through ethers.js, viem, or web3.py.
Observability and Alerting
The observability stack combines industry standard metrics and logging infrastructure with blockchain specific tooling. Grafana dashboards visualize agent performance and system health. Prometheus collects time series metrics. PagerDuty manages on call escalation. Slack integrations provide team visibility into routine operations and anomalies.
Adjacent work we do.
Governance & treasury
The decision layer that defines the rules agents operate within, and the treasury systems they manage.
Smart contract systems
The scoped roles, access controls, and timelocks that keep every automated action inside a known envelope.
Crypto payment solutions
Settlement automation, payment routing, and the operational infrastructure where keepers deliver the most immediate value.
Tell us what your operations need to survive at scale.
Describe the jobs that wake someone up at night, the manual processes that bottleneck your team, or the on chain conditions that need faster response. We will scope the automation that fits.