A tool chain into a production database
A LangChain agent follows a chain of steps into a production database it was only ever meant to read from a test copy of, and issues a write against live data before anyone reviews the chain.
Use case · Developers & SMB
Governing LangChain agents
Tool-calling LangChain agents can reach databases, APIs, and shells. Bound exactly what they touch, cap what they spend, and record what they did.
The risk
What an ungoverned agent gets wrong here.
LangChain (the popular open framework for building tool-using AI agents) makes it easy to give an agent a set of tools and let it decide how to chain them. That is the power, and the risk. A tool-calling agent can reach a database, call an external API, or run a shell command, and it chooses which to call based on a prompt and its own reasoning. The tools you registered define what it can do; nothing outside your code defines what it should be allowed to do right now. A prompt injection in a document, a bad chain of reasoning, or a tool that returns something unexpected, and the agent takes an action against a real system that you never intended. Agentomy sits in front of the tools a LangChain agent calls and governs each action as it happens.
Without governance01 02 03 04
Where it goes wrong.
An injected instruction hijacks a tool call
A document or a tool response the agent reads contains a hidden instruction telling it to call a different tool or reach a different system. The agent obeys, and a tool it should never have used in that moment fires.
A shell tool that runs the wrong command
An agent with a shell tool runs a command well outside its task, a deletion or an unexpected network call, because the tool was available and nothing scoped when it could be used.
An API tool that runs up cost
A tool that calls a paid API gets invoked in a loop, and the agent's own reasoning keeps it going, spending far past what the task was worth with nothing capping the rate.
With Agentomy01 02 03 04
How Agentomy governs it.
Discover the tools and reach of each agent
See every LangChain agent you run and the tools, keys, and systems each one can reach, so the actual reach of a tool-calling agent is visible rather than buried in your code.
Authorize every tool call before it runs
Each action a tool would take is checked against what the agent is allowed to do before it executes. A read-only agent cannot write, a database tool cannot reach production, and a shell tool cannot run a command outside its scope. Permitted calls pass, the rest are refused and logged.
Cap spend and rate, and halt on demand
Hard rate and spend limits stop a tool-calling loop from running up cost, and one command halts every governed agent immediately when a chain goes somewhere it should not.
Record every tool call
A tamper-evident, hash-linked log captures every tool call an agent made (which tool, against which system, and why), so you can trace exactly what a chain did after the fact.
See it in the record
Every action, logged and provable.
A tamper-evident, hash-linked trail of every governance decision for this workload: what an agent did, under whose authorization, and why. Plain-English reasons for every allow and deny, exportable to the framework your auditors care about.
Agentomy Command CenterFilter Governing LangChain agents
Fleet posture
- Governance posture77/100
- Agents governed5/ 6
- Quarantined1
Illustrative control mapping
- EU AI Act78%
- NIST RMF92%
- OWASP Agentic85%
Recent activity
- 12:42PermissionRouterVendor access blocked
- 12:41AuditLoggerEvidence record generated
- 12:40RuntimeMonitorBehavior drift flagged
- 12:39TrustScorerGovernance score updated
- 6 Registered
- 1 Quarantined
- Fleet Active
- Pipeline 10 Stages
| Agent | Type | Status | Module | Score |
|---|---|---|---|---|
| process-discovery-agent | Discovery | Active | RuntimeMonitor | 92 |
| exception-triage-agent | Exceptions | Active | AuditLogger | 88 |
| vendor-review-agent | Procurement | Monitored | PermissionRouter | 74 |
| invoice-extract-agent | Finance | Quarantined | HaltProtocol | 41 |
| runbook-draft-agent | Runbooks | Monitored | EthicsConstraint | 79 |
| throughput-report-agent | Reporting | Active | DecisionLog | 90 |
Live 12:42:08Governance decisions, newest first
| Agent | Action | Policy | Result | Time |
|---|---|---|---|---|
| PermissionRouter | data_access_request | vendor-access | Blocked | 12:42:08 |
| AuditLogger | output_validation | audit-trail | Allowed | 12:41:54 |
| TrustScorer | policy_check | runtime-score | Allowed | 12:41:37 |
| RuntimeMonitor | behavior_drift | drift-threshold | Flagged | 12:40:58 |
| HaltProtocol | halt_initiated | unsafe-action | Blocked | 12:40:21 |
| EthicsConstraint | prompt_review | restricted-output | Flagged | 12:39:46 |
| DecisionLog | evidence_recorded | decision-proof | Allowed | 12:39:12 |
- 100% Integrity
- 43,482 Blocks
- SHA-256 hash-linked
| Block | Timestamp | Agent | Action | Tier | Hash |
|---|---|---|---|---|---|
| 43,482 | Today 12:42:08 | process-discovery-agent | data_access_request | Evaluator | 9379d1d35b670693 |
| 43,481 | Today 12:41:54 | exception-triage-agent | output_validation | Analyst | e57e4e5d05766c2c |
| 43,480 | Today 12:41:37 | vendor-review-agent | policy_check | Builder | 59d19fb8b2c71755 |
| 43,479 | Today 12:40:58 | invoice-extract-agent | behavior_drift | Operator | b8e4e0acc90eecc0 |
| 43,478 | Today 12:40:21 | runbook-draft-agent | halt_initiated | Strategist | c4f71e4ddd8c9bd6 |
| 43,477 | Today 12:39:46 | throughput-report-agent | prompt_review | Evaluator | a9e621e8134d7ded |
Governance events today: 12,842Demo environment