Bring your own AI agent
Use Claude, ChatGPT, Codex, Copilot, Gemini, or a custom runtime. ITMS does not require one model vendor.
The AI tax API for reviewable, AI-agent-driven work.
Your AI agent can work with Prophit.ai’s AI agent through governed interfaces. Connect the runtime your team already uses to documented ITMS operations: your AI agent prepares requests and delegates supported tax work, and ITMS answers with evidence-backed, reviewable results instead of pretending a person clicked through software.
AUTHORIZED AI AGENT WORKFLOW
1 read ITMS documentation2 select supported endpoint3 submit scoped request4 inspect result + evidence5 escalate human decisionHuman authority remains at the action boundary.Call governed tax workflows
Your AI agent works through documented interfaces, schemas, credentials, permissions, and predictable error states.
Keep the work inspectable
Requests, responses, evidence, exceptions, and authority boundaries remain visible to the people accountable for the result.
What an AI agent can actually do here.
Concrete workflows grounded in the shipped product modules. Each one returns reviewable results, and each one stops at the authority boundary that protects your operation.
Handle this period’s filing workflow
Your AI agent gathers or prepares transaction inputs, runs supported calculation, reconciles expected against collected tax, surfaces exceptions, prepares filing-ready outputs, and requests your approval.
A supported filing lane is used only when it is configured and separately authorized.Explore Sales Tax Compliance →Analyze nexus exposure
It reviews economic and physical-presence signals, approaching thresholds, and obligation status across states, then reports where filing work may be required.
Registration is not supported in ITMS; that decision and step stay with your team.Explore Sales Tax Compliance →Review vendor-charged tax
It compares tax charged by vendors against expected tax with reason-code support, ranks variances by materiality, and prepares use tax accrual and overpayment candidates for review.
Reviewers adopt or correct candidates; the AI agent does not finalize accruals on its own.Explore Purchasing Tax Compliance →Renew expiring certificates
It identifies expired and expiring exemption certificates against state legal rules and your internal policies, then prepares supported renewal outreach for the affected customers.
Outreach is sent only with configured communications authority.Explore Exemption Certificate Management →Recover incorrectly taxed purchases
It reviews purchasing or audit results, assembles evidence-linked overpayment findings, and prepares or issues supported vendor exemption certificates for purchases that were taxed in error.
Findings carry their evidence; no refund, authority, or legal outcome is guaranteed.Explore Risk & Recovery →Prepare a cross-system tie-out
It lines up billing, ERP or GL, tax engine, and filed-return totals, computes deltas, drafts explanations for material exceptions, and stages the run for reviewer signoff.
Prepared, reviewed, and approved states stay with people; unsafe actions remain blocked.Explore Reconciliation →Explain the period at a glance
It pulls entitlement-scoped analytics for the active organization and location, then summarizes exposure, readiness, and throughput with links back to the underlying work.
Summaries cite the operational data they came from instead of inventing totals.Explore Analytics & Reporting →Implemented today. Headed toward agent-to-agent.
Be clear-eyed about which part is shipped and which part is direction. The governance model is the same in both.
A documented API your AI agent can already use
- Documented HTTP interfaces, schemas, errors, and rate limits
- Starter packs for Claude, ChatGPT, Codex, and any custom runtime
- Scoped credentials, sandbox access, logging, and throttling
- Deterministic calculation that fails closed, plus configured AI-assisted review paths
- Evidence and audit context returned with the work
Delegate whole workflows between AI agents
Over time, your AI agent should be able to hand Prophit.ai’s AI agent a whole outcome, such as preparing a period for filing review, and get back finished, evidence-backed work rather than orchestrating each call itself.
The boundaries do not loosen as that surface widens: the same plans, configuration, jurisdiction coverage, scoped credentials, evidence, and human approval points govern every delegated step. New connection surfaces will appear in the API documentation when they ship, not before.
AI-agent-ready across the compliance lifecycle.
Start with one bounded operation, then expand only where your plan, configuration, jurisdiction coverage, credentials, and review model support it.
Organization and location setup
Create the operating context an authorized workflow needs before tax work begins.
Data intake and validation
Move structured transaction data into supported workflows, surface missing fields, and inspect validation results.
Calculation and classification
Use deterministic tax rules where coverage exists and controlled assistance where a reviewable classification is appropriate.
Certificates and exemptions
Connect supported certificate records, exemption status, and transaction context without hiding the source evidence.
Reconciliation and exceptions
Compare source and tax records, identify variances, and route unresolved questions to the right reviewer.
Returns preparation
Organize reviewed period data and filing evidence while keeping submission, payment, and approval authority explicit.
Deterministic when unknown must mean stop. AI-assisted when review can absorb uncertainty.
Deterministic
Apply supported rules and fail closed when coverage is unavailable. A missing rule does not become a guessed tax result.
AI-assisted
Use configured probabilistic assistance for eligible unknown classifications, keep the result reviewable, and preserve human escalation.
What an AI agent workflow actually does.
The useful distinction is not whether AI is present. It is whether each step has the context, interface, evidence, and authority it needs.
- 01
Discover
Read the current documentation, schemas, supported operations, and environment requirements.
- 02
Prepare
Gather the organization, location, transaction, certificate, or period context the operation requires.
- 03
Invoke
Send a scoped request with the right credential and stop when validation or coverage is unavailable.
- 04
Review
Inspect the result, evidence, warnings, exceptions, and audit context before relying on the output.
- 05
Act with authority
Keep filing, payment, approval, and production actions behind the separate human or system authorization they require.
Claude, ChatGPT, Codex, Gemini, Copilot, or your own AI agent runtime.
ITMS is built around documented HTTP interfaces rather than a single model vendor. Download a branded starter pack for your AI agent harness, then review its authority limits before connecting credentials.
Download .md
Universal starter pack
Works with any AI agent runtime
Get the pack ↓Download .mdClaude skill
Claude projects and AI agent workflows
Get the pack ↓Download .mdChatGPT instructions
Custom GPTs and AI agent workflows
Get the pack ↓Download .mdCodex skill
Repository and integration work
Get the pack ↓Gemini, Copilot, and custom runtimes can start with the universal pack. The live API documentation remains authoritative.
See the path before you connect credentials.
This short local product walkthrough moves from the agent-to-agent overview through supported surfaces, deterministic and AI-assisted modes, starter packs, and the three-step authority checklist.
Three steps to a bounded proof of concept.
- 01
Read the quickstart and schemas
Choose a supported workflow and understand its request, response, error, and rate-limit behavior.
- 02
Use scoped credentials in the right environment
Keep secrets out of prompts, files, and logs. Limit the AI agent to the organization and operations it needs.
- 03
Review before expanding authority
Inspect results and audit logs before adding broader operations or any downstream filing action.
Questions a responsible AI agent program should answer first.
Is this a chatbot?
No. The AI agent experience is a way for authorized software to work through supported ITMS operations. Natural-language reasoning can help prepare and interpret work, while the tax operation still uses documented interfaces and controls.
Does the model calculate tax by guessing?
No. Supported deterministic rules remain the source of tax outcomes where unknown information must cause a stop. AI assistance is reserved for configured tasks where uncertainty can be disclosed and reviewed.
Do we have to replace our current AI stack?
No. Start with the runtime your team already governs. Use the universal pack or a platform-specific starter file, then connect only the credentials and operations needed for the first workflow.
Can an AI agent file or pay automatically?
Only when the product surface, entitlement, environment, and separate action authority support it. Reading documentation or calculating tax does not silently grant filing, payment, approval, or production authority.
Do you support MCP or an agent-to-agent protocol?
The supported integration surface today is the documented HTTP API plus the starter packs on this page. We do not claim protocol support we have not shipped; when a connection surface is added, it will appear in the API documentation with the same authority model.
Can an AI agent register us in new states?
No. Registration is not part of ITMS. An AI agent can analyze nexus exposure and approaching thresholds so your team can decide what to do about them.
Bring us the workflow, not just the model name.
We’ll map the data, endpoints, deterministic rules, AI-assisted steps, review points, and authority limits needed for a responsible integration.
- Production standalone Sales Tax and Use Tax Engines ($99/month per location through sales-assisted setup, with 10,000 calls/month included) return determinations to source systems through an API.
- The tax content is published and auditable: the Sales Tax Atlas carries 51 state and DC guides and 3,800+ county and city rate pages, each pinned to its official source.