Every action an AI agent takes should be explainable after the fact.
Decision Horizon is building the system of record for AI agents: a signed account of what an agent knew, what it was allowed to do, and what it did, written at the moment it acted.
Agents are becoming actors, not tools.
They read customer records, change code, move money and file tickets with real credentials. Anything that acts on a company's behalf will be asked to account for what it did.
Logs are not evidence.
A log written by the system under review proves little to someone who doesn't trust that system. Evidence has to be written before the action, outside the agent, and checkable by anyone with a public key.
Honest records beat confident ones.
A record that guesses is worthless the moment one guess is wrong. Ours says "unknown" when it can't know, and lists its own blind spots, so the rest of it can be trusted.
Companies are handing agents the keys to production faster than they can answer basic questions about them. When something goes wrong, the questions are always the same: what did it know, what was it allowed to do, and why did it act? Today the answer is usually a log file and a lot of reconstruction.
We think that has to change before agents can be trusted with the work people want to give them. So we built the layer that sits on the agent's tool path, decides each consequential action against policy, and leaves a signed record behind. The record works for the security team in the moment, and for the auditor, the regulator or the court months later.
Built by an engineer who hit the problem first.
Hans Ade
Founder & CEO
Hans builds Decision Horizon end to end: the record format, the MCP proxy, the model-API gateway, the Python and TypeScript SDKs, the verifier and the hosted service.
Before Decision Horizon, he founded PiPilot, an AI-native development platform whose coding agents work inside users' own GitHub, Vercel and Supabase accounts. Running those agents with real permissions is where he first needed proof of what an agent knew and was allowed to do, and found only logs. PiPilot is now Decision Horizon's Customer Zero.
A self-taught engineer based in Douala, Cameroon, Hans graduated from Founder Institute New York (Spring 2026) with PiPilot. His first win was first place at the 2022 Littoral Regional Tech Hackathon.
Working in the open on how agent evidence should work.
Cloud Security Alliance
Members of the AI Safety and AARM working groups, contributing to guidance on evidence for agent actions.
NIST CAISI
Submitted a practitioner case study on monitoring agent access to NIST's Center for AI Standards and Innovation.
Open verification
The verifier is public and runs offline. Anyone can check a Decision Horizon record without asking us, which is the point.
State the limit.
Every capability ships with what it can't do, written down where users will see it. If we find a gap, we fix it or we publish it. Ten review rounds produced 124 findings, and each went one of those two ways.
Show, don't claim.
We'd rather put a measured number, a test or a verifiable record in front of you than an adjective. No customer logos we haven't earned, no testimonials we didn't get.
Trust no one, including us.
We design so that you never have to take our word for it. Your keys stay yours, and every record can be verified without us.
Open roles
We're a small founding team, and every early hire shapes the product and the company. Roles are remote, with founding equity. We'll talk openly about salary and stage in the first conversation.
Don't see your role? Write to hans@decisionhorizon.dev with what you'd want to work on.
Co-founder, Go-to-MarketLeadership · Remote · Full-time
Own how Decision Horizon reaches the security, risk and compliance teams who need it. You'll turn design-partner conversations into paid pilots, and paid pilots into a repeatable motion, as a co-founder with a co-founder's stake.
What you'll do
- Run the pilot pipeline end to end, from first conversation to signed order form.
- Sell to CISOs, heads of risk and AI platform leads at fintechs, insurers, health companies and agent platforms.
- Shape pricing, packaging and the pilot offer with what buyers actually say.
- Build the relationships with auditors and standards bodies that make the record the expected answer.
You might be a fit if
- You've sold security, compliance or infrastructure software to enterprises, ideally into regulated industries.
- You can hold a technical conversation with a security architect and a commercial one with a CFO.
- You'd rather build the first sales process than inherit one.
- You want founder-level ownership and equity.
Founding Engineer, Security & CryptographyEngineering · Remote · Full-time
Harden the core of the product: the record format, signing, hash chains, witnesses and the verifier. Your work is the part an adversary and an auditor will both read closely.
What you'll do
- Own the signing, key custody and witness-quorum code paths, in Python and TypeScript.
- Run and extend the adversarial review process, and publish what it finds.
- Prepare the product for a third-party penetration test and SOC 2.
- Work with design partners' security teams on threat models for their agent deployments.
You might be a fit if
- You've shipped security-critical code: cryptographic protocols, transparency logs, authentication or tamper-evident storage.
- You think in attacker capabilities and can explain a trust model plainly.
- You're comfortable in Python, and in TypeScript or Rust.
- You care about stating limits honestly.
Founding Engineer, Platform & IntegrationsEngineering · Remote · Full-time
Make Decision Horizon drop into every agent stack. You'll build integrations, the hosted record store and the Console, and keep the insertion points fast and boring.
What you'll do
- Extend the MCP proxy, model-API gateway and SDKs to the frameworks and platforms customers use.
- Build and run the hosted store, API and Console: reliability, tenancy, retention and export.
- Integrate with SIEMs, approval channels (Slack, ticketing) and identity providers.
- Keep latency and footprint measured and published.
You might be a fit if
- You've built developer infrastructure, API gateways or observability pipelines in production.
- You're strong in TypeScript/Node and Python, and at home with HTTP, streaming and protocol details.
- You've worked with LLM agents, MCP or tool-calling frameworks, or want to go deep on them.
- You like writing documentation as much as code.
Forward-Deployed EngineerCustomer engineering · Remote · Full-time, with travel
Be the person who makes a pilot succeed. You'll install Decision Horizon in a customer's own environment, write their first policies, and sit with their auditors when the replay is reviewed.
What you'll do
- Lead technical onboarding for pilots, on the customer's infrastructure.
- Write policies and approval flows for real agent workflows in finance, insurance and health.
- Turn what you learn in the field into product changes and documentation.
- Run demos and technical deep-dives with prospects.
You might be a fit if
- You've been a solutions, sales or forward-deployed engineer for a security or developer-infrastructure product.
- You can read a customer's architecture quickly and find where the agent's tool path really runs.
- You explain technical trade-offs clearly to non-engineers.
- You're comfortable working across time zones with US and European customers.
Frequently asked
What does Decision Horizon actually do?
It sits on an AI agent's tool path: in front of an MCP server, in front of an OpenAI-compatible model API, or around a function. Before each consequential action it checks policy and writes a signed record of who sponsored the action, what authority it held, what the agent had read, which rule decided and what the record can't see. After the action it writes the outcome.
How is this different from observability or tracing?
Tracing tools help developers debug, and their data is written and kept by the system being observed. Decision Horizon decides before the action runs and produces evidence that a third party can verify offline with a public key.
Do we have to trust you with our data?
No. Records are written on your own hosts by default and signed with your key. If you use the hosted store, it holds only your public keys, and every chain can be downloaded and verified without us.
Does it change how our agent behaves?
Only where policy says so. The proxy passes the same tool list through unchanged. Allowed calls go through as before; blocked calls are refused, and checkpoints wait for a person.
How do we start?
Pick one agent workflow that matters. We put Decision Horizon in front of it with you, and replay a real session's record together. The quickstart shows how to try it yourself first.
Talk to the person who wrote the code.
Bring one of your agent workflows. Or write to hans@decisionhorizon.dev.