Stategram
Stategram provides internal controls for AI agents that move money, issuing action-specific permits instead of blanket API keys. The platform enforces per-action limits and cross-agent daily totals, routing anything above signed thresholds to named human approvers. Permits are cryptographically signed in secure enclaves and verifiable offline in about 10 ms.
- Regulatory & Compliance Technology
- Artificial Intelligence
- Cybersecurity
- AI Agents
- Enterprise Software
- Developer Tools
- Software Only
Funding
Founders
Founder details are not available yet.
Product
Problem
Organizations deploying AI agents face a governance gap: standard API keys authenticate the caller but do not constrain what a single action may do, and system prompts are vulnerable to injection attacks that can trigger harmful actions like unauthorized refunds or data changes. One wrong tool call in production can result in significant financial loss and reputational damage.
Solution
Stategram provides maker-checker controls for AI agents by issuing action-specific permits instead of blanket API keys. Each permit defines what one action may do under limits signed by risk owners, and any action above those limits routes to a named human approver. The platform enforces cross-agent daily totals that aggregate activity across every model and human, preventing attacks that slip under per-action thresholds. Every permit and referral is cryptographically signed inside a secure enclave and written to an immutable log, producing receipts that auditors can verify offline without trusting Stategram or the customer's infrastructure.
Target Audience
Primary users are risk, compliance, and platform engineering teams at enterprises deploying AI agents for consequential actions like refunds, proposals, or deployments, particularly in regulated industries subject to EU AI Act human-oversight and record-keeping obligations.
Features
- Action-specific permits bound to exact payloads and policy versions, with single-use, short-lived validity
- Named approver routing for actions exceeding signed limits, with approval through Slack or Teams
- Cross-agent daily totals covering every agent, model, and human to stop drip attacks below individual thresholds
- Cryptographic signing inside AWS Nitro Enclaves with KMS-held keys and a write-once S3 Object Lock log
- Offline receipt verification in about 10 ms via an MIT-licensed checker that requires no Stategram dependency
- Integrations with LangChain, LangGraph, CrewAI, MCP, Claude, ChatGPT, and Slack; deployment in customer tenant or EU-hosted (Frankfurt) infrastructure
- Human-readable policy language that risk officers can read and sign without engineering support
- Refusals return the specific rule that fired, providing clear audit evidence for every blocked action