Software services · capability-led engagements

Make the evidence behind an AI result reviewable.

Maha helps AI teams prepare source-linked inputs, measure evidence retention, and build handoffs that show what was checked and what still needs judgment.

Bring one recurring workflow and a representative sanitized example. We will identify the smallest useful component, the checks it can perform, and the evidence needed to decide whether it belongs in your system.

Preserve the evidence your agent needs

For engineers supplying documents to AI agents.

A budgeted context pack, source and passage identifiers, hashes, and an exact report of which caller-labelled evidence spans survived selection.

Example: Prepare a project brief while retaining the owner-approved constraints and recording any omissions.

Existing API contracts; check deployment availability and account entitlement.

Read the API contracts

Evaluate a context pipeline on your workload

For product and engineering leads deciding whether to adopt context infrastructure.

A frozen workload comparison, per-case findings and a proceed, revise or stop recommendation.

Example: Compare your existing retrieval or gateway controls with Maha on the same source set and required evidence.

Fixed-scope assessment; scope and data handling agreed before work starts.

See assessment scope and pricing

Build a reviewable context handoff

For agent-platform teams handing persistent context to a reviewer or another tool.

A local workflow that binds review to source revisions, saves approval and replay state across restarts, and writes a source-linked review artifact.

Example: Hold a write when evidence changes after approval, then reconcile an interrupted write before considering another attempt.

Local evaluation implementation. Remote Fulcra transport and production operation are not enabled.

Inspect the handoff workflow

Capability and procurement facts

Choose by the evidence and data boundary of the specific product. The answers below distinguish demonstrated checks from requirements an integration still needs to satisfy.

What does the software check?

Context evaluation measures exact retention of supplied source spans. The handoff also checks revision bindings, policy, approval and replay state.

Factual truth, complete evidence discovery, general entailment and domain suitability require separate evidence.

Evaluation contract

Does it call external AI providers?

The deterministic local compiler and handoff demonstration run without a model call. Other products, including model-backed assessments, have their own provider paths.

There is no company-wide zero-external-API promise.

Assessment and security scope

Where is information retained?

The local handoff file store contains source text, and plans/artifacts include selected passages. Its separate gateway database retains approval, digest and dispatch metadata.

The host controls access, backups and deletion. A 24-hour universal deletion guarantee is not offered.

Local data boundary

Can we rely on certification or SSO?

This local workflow has no independent security certification established by the reviewed evidence. Host identity and reviewer authentication must be supplied by the integrating system.

SOC 2 Type II, SAML/OIDC and a production SLA are not represented as included capabilities.

Integration limits

Can it safely resume interrupted work?

The local SQLite adapter commits an intent before writing. Reopened or concurrent workers return the stored state; unknown outcomes require reconciliation.

This is not a distributed exactly-once guarantee. Remote consistency, authenticated identities and operational review remain release gates.

Workflow and acceptance checks

What do we need to provide?

One recurring task, permitted source data, reviewed revision/evidence requirements, an expected output, a baseline and an authorised reviewer.

Start with synthetic or public data. Private or regulated workloads require separately reviewed controls.

Intake checklist

Start with one workflow

To assess fit, send a short description covering these points. Leave credentials, personal data and confidential source documents out of the first enquiry.

  1. The recurring decision or handoff and the person responsible for accepting its result.
  2. One representative synthetic/public packet, source revisions and the evidence that must survive.
  3. Your current workflow and review process, including existing safeguards.
  4. The intended destination/tool, permitted effect and reviewer authority.
  5. The result that would make the experiment worth repeating, plus any relevant budget or deadline.
Discuss a software workflow

What the result means

A matching quote establishes that text was found. A retention check establishes that a declared span survived selection. A hash commits to bytes. These checks support review; factual accuracy, completeness and authority require their own evidence.

Our API also includes domain-specific computations and licensed research content. Each has its own contract and limits. The SDK is an integration tool; payment, technical completion and customer acceptance are separate states.

Browse the wider catalogue and purchasing options