chainforeFree discovery call

Permission to act. Proof that it happened.

Chainfore is an advanced AI systems engineering company with an applied research programme. We build AI systems that act only within the permissions you grant, confirm each outcome in the system of record, and recover when a step fails.

Free discovery callOur research programme

For growing companies, mid-market companies and enterprises.

Work delivered for

  • PreScouter
  • PepsiCo
  • Sony with PreScouter
  • Saudi Aramco with PreScouter
  • Hitachi Rail
  • Emergent BioSolutions

Sony and Saudi Aramco: private-network AI deployments, delivered with PreScouter.

Our thesis

Reliable autonomy. Useful delegation to AI, within limits you can inspect.

  1. 01
    The right contextPermitted information reaches the workflow with its source, freshness and access rights attached.
  2. 02
    Authorised actionsThe system acts only within explicit permissions. Consequential actions wait for a person to approve them.
  3. 03
    Checks on actual outcomesSuccess is confirmed by reading the system of record after the action.
  4. 04
    Recoverable executionCheckpoints, retries that cannot duplicate a transaction, and a way back after a failed step.
  5. 05
    Accountable ownershipA named person owns each decision to widen what the system may do.

The failures we test for

A confident explanation is not a verifier. So every system is tested against the failures that matter.

The model reports success, but nothing changed. The check reads the system of record, and the false claim fails.

An approved action is altered before it runs. Approval is bound to the exact parameters and an expiry, so the altered action is blocked.

A timeout hits after a write and the step retries. Idempotency keys make sure the transaction is never duplicated.

A retrieved document tells the agent to do something else. Retrieved content is treated as data, and the instruction is refused.

Applied research

Reliable Enterprise Autonomy. One research programme, four tracks, each with its own question.

  1. A
    Agent reliability and evaluationDo outcome verifiers reduce false completion?
  2. B
    Context, memory and permissionsDoes permission-aware, freshness-checked context outperform a larger unfiltered one?
  3. C
    Execution and recoveryWhich recovery mechanisms matter most when tools fail?
  4. D
    Economics and model selectionWhen does a smaller model suffice, and what does an accepted outcome cost?

Our publication standard: each result states its question, baselines, data, limits and negative results, and is reviewed by someone who did not build the system.

Research programme

Offline. Shadow. Supervised. Limited. Autonomy widens one level at a time, on new evidence.

  1. 1
    OfflineTested against synthetic fixtures.
  2. 2
    ShadowObserves real work without acting.
  3. 3
    SupervisedAsks for approval before each bounded write.
  4. 4
    Limited autonomyCarries out explicitly low-risk actions within validated limits.

Scope expands only after the owner approves and new evidence is produced. A change to tools, model version or permissions restarts the relevant tests.

In production

Auxee, for PreScouter. A market-intelligence analyst inside Microsoft Teams.

PreScouter’s team asks Auxee market-intelligence questions where they already work, in Microsoft Teams. Chainfore built it on Activepieces, and it is live in production.

Client
PreScouter
Built on
Activepieces

We also delivered private-network AI deployments for Sony and Saudi Aramco with PreScouter, and AI work for PepsiCo from 2025 to February 2026.

Client work

Working with Chainfore

Five engagements. Each with a written scope, a named owner and acceptance criteria, agreed before work begins.

  1. 01Workflow architecture and baselineDecide whether to build2–3 weeks
  2. 02Assured workflow engineeringBuild one workflow6–10 weeks
  3. 03Agent release assuranceTest an agent you already have4–6 weeks
  4. 04Reliability operationsKeep a live workflow dependableMonthly
  5. 05Dedicated engineering deliveryHand over an engineering workstream3–24 months

For growing companies, mid-market companies and enterprises.

Define. Bound. Engineer. Verify. Operate. The same five stages in every engagement, each with a named output.

  1. 01
    DefineThe workflow owner, the current process and the acceptance criteria. AI is compared with simpler options first.
  2. 02
    BoundData access, allowed tools, actions that need approval and the risk owner, written down before the build.
  3. 03
    EngineerIntegrations, state, context handling and recovery behaviour.
  4. 04
    VerifyNormal, hard and adversarial cases, checked against the resulting state of the systems involved.
  5. 05
    OperateDeployment, monitoring, change control and a handover your team can run.

How we deliver

Trust, stated as status. What is in place, what is in progress, and what is not held.

  • In place

    Delivery method, failure tests and evaluation standard, applied to each engagement. A data-flow diagram, threat model and model and provider register for each engagement. Confidentiality and IP terms in every contract. No training on your data unless you authorise it in writing.

  • In progress

    A standard security pack for procurement review. The ISO/IEC 42001 certification pathway with an accredited body.

  • Not held

    SOC 2. We will claim it only when an independent auditor issues a report.

Security and dataTrust and compliance

Haseeb JavedFounder and Chief Executive Officer

Haseeb leads Chainfore’s direction, research agenda, technical leadership and client relationships. His background is in enterprise technology and software engineering, including Accenture and PwC.

Contact

Bring one workflow. Important enough to fund, not yet dependable enough to deploy.

Tell us the process, its current baseline, the systems involved and the cost of failure. We reply with a proposed first step.

Free discovery callsales@chainfore.com
Add more detail (optional)
We reply with a proposed first step.