Try Token Observe free for 30 days.
Create a cloud workspace for £0 for 30 days. No card or automatic paid conversion. Customer-hosted licensing and assisted pilots are quoted separately.
The hosted service is a small cloud pilot, not broad general availability. Its admission, process, storage and traffic limits apply across the service. Customer-hosted design-partner arrangements are available separately.
Cloud trial and customer-hosted options.
See the bill, itemised.
Track connected provider-account costs and the AI subscriptions and API accounts you register.
Keep a record of what it was asked to do.
Follow requests routed through your workspace. Bounded, redacted text capture is off by default.
Set the rules for the agents you connect.
Record tool approvals and govern permissions and budgets for agents you connect.
Free 30-day cloud trial
£0 trial feethirty days from workspace creation, with no card or automatic paid conversion
Who it is for. The IT team connecting a supported provider billing source and evaluating optional agent governance.
Create an account, confirm email and create a hosted workspace. Connect provider-reported billing with the required reporting credential. Agent activity requires a separate inference key, a registered agent and supported client configuration.
- A hosted workspace; no Token Observe server installation
- Billing adapters for OpenAI, Anthropic Claude, xAI Grok and Google Cloud Billing
- Source periods, currencies, collection freshness and partial/failure status
- Governance and activity for requests routed through the workspace
- Provider model calls, model-assisted search and Google BigQuery charges
- Historical prompts, browser conversations and all subscription-plan bills
- An availability SLA, independent penetration test or production-critical assurance
- Unlimited admission, storage, running workspaces or traffic
Self-hosted deployment subscription
On applicationper deployment, per term, with a ceiling on simultaneously active agents
Who it is for. The head of IT who owns the AI line on the card statement, and the team who would run the deployment.
License one customer-hosted installation with an agreed ceiling on active agents. Commercial capacity checks apply when configuration changes; they do not stop already governed traffic. Review the complete licence and operating requirements before signing.
- One deployment on your own infrastructure. Air-gapped is granted
- Source access: read, compile, modify, own your separable modifications
- A reasonable number of non-production copies, counting towards no limit
- Updates during the term, on a pre-1.0 policy: the latest published minor only
- Every governance control, in every tier. A licence scales capacity, never enforcement
- Model spend. Your keys, your provider’s invoice, no margin taken
- Your infrastructure: hosts, network, TLS, secrets, DNS, disk, backups
- Model behaviour and provider incidents. Configuring failover is covered
- Your agents’ own code. Reading a trace with you is covered
- Any availability or performance commitment (section 8.2)
Design-partner agreement
On applicationstated on one design-partner Order Form, with the term and the scope
Who it is for. An organisation that wants a named engineer and real influence over the roadmap, and accepts a pre-1.0 product’s risks in writing.
Work directly with a named engineer and help shape the roadmap. The agreement asks for regular feedback, redacted diagnostics when you raise an issue, and access to any penetration test you commission. Publication rights remain as set out in the licence; there is no support desk or service-credit schedule.
- A named engineer, reachable directly, who has read your deployment
- The response targets below, written into the agreement rather than a web page
- Direct influence on the roadmap, with your blocking issues prioritised explicitly
- Early access to fixes on a branch, and full sight of the defect list
- Service credits. There is no availability SLA to credit against
- Cover outside 09:00–17:30 Europe/London, Monday to Friday, or on English public holidays
- Known issues already published, and defects arising from your own modifications
- For customer hosting, data recovery: there is no vendor-side copy of your database
Policy and control-framework engagement
On applicationquoted per engagement, against a written scope
Who it is for. A team commissioning policy authoring or control-framework mapping.
Commission policy authoring and framework mapping against an agreed written scope. Policy design guidance and template review remain part of support; separate authoring or compliance-function work is quoted as professional services.
- Authoring your policy set, with the shadow-mode backtest a named person can be required to acknowledge before enforcement
- Mapping it onto your control framework, and onto the EU AI Act, ISO/IEC 42001, NIST AI RMF and OWASP
- Compliance approval or sign-off of your DPIA, FRIA, ROPA or risk register
- Certification or an independent attestation
- Investigation or remediation of agents found by the radar
Thirty days in a hosted workspace.
The hosted trial lasts thirty days from workspace creation, with no card or automatic paid conversion.
Billed subscriptions, billed API charges and the recorded API cost estimate: three figures on one screen with no total across them. The product's own banner says why: gateway costs and API invoices are separate views of potentially overlapping usage and are not added together.
Sample data in the real Token Observe interface, recorded from a local evaluation with a temporary database, a mock provider and one request. No live provider key was used. The product stamps its own demo rows SYNTHETIC EVALUATION DATA in these pixels.
- 01Open the account form, confirm your email using its link, then sign in and create a workspace. No ZIP, Node.js, Docker or local Token Observe server is required.
- 02Connect OpenAI organization costs with an Admin API key; Anthropic organization costs with its Admin API key; xAI team API usage with a management key and team ID; or a configured Google Cloud Billing BigQuery export with its required reporting access. These sources do not cover every subscription or plan.
- 03The report covers the last 30 complete UTC days and preserves currency, source periods and collection status. A completed collection does not mean the provider's bill is final. Missing or failed evidence is unknown, not zero.
- 04Billing credentials are separate from inference keys. Model calls and model-assisted search use your separate inference account and can incur provider charges. Google BigQuery scans can also incur charges; saving an inference key does not make a model call.
- 05At trial expiry, workspace changes, model calls, scheduled billing collection and paid-capable natural-language trace searches stop. Authorized retained GET reads, exports and exact first-party MCP reporting reads remain subject to the shared monthly traffic quota.
- 06The small pilot has shared admission, process, storage and traffic limits. Reaching the traffic quota pauses portal and gateway access until an operator raises it or the UTC month renews; retained data is not erased by that quota.
- 07Cloud hosting holds workspace evidence and supplied credentials on the service. Workspace processes have separate directories and databases but share one OS user and host; this is not a sandbox or VM security boundary.
- 08There is no availability SLA or independent penetration test. Keep production-critical workloads outside this pilot. Customer-hosted installations have separate licence and operating terms.
A separate source-available licence.
- Use, modify and self-host under a licence
- Read the entire governance domain before buying: the core package is pure functions with zero runtime dependencies
- Inspect, test, fuzz, penetration-test and reverse-engineer your own deployment, or commission a third party to do it for you
- Publish benchmark results with no pre-approval, provided the version and configuration tested are named; publish security findings after coordinated disclosure
- Redistribution
- Offering Token Observe as a competing hosted service
The published customer-hosted licence is a draft requiring legal review. Your signed agreement sets the final terms.
Ask James for the licence, current defect list and customer-hosted support terms. Source access is arranged separately from the cloud trial. Request review materials
Evaluate, scope, quote, sign.
Paid cloud pricing is not published. The free trial ends without an automatic upgrade. Discuss customer-hosted licensing, support or professional services against a written scope before agreeing a quote.
Start the free trial
Create an account, confirm email and create a cloud workspace. The 30 days start at workspace creation.
Send the shape of your estate
Deployments, agents active at once, providers. All optional.
A quotation against a written scope
A quote tied to the deployment, agents and support you need.
An Order Form
The term, the fees and the permitted scope. Model spend never appears on it.
Only the last step commits you to anything.
Who the pilot suits.
- A team that can connect a supported provider billing account and check its reporting coverage.
- A team that wants to add agent governance after configuring a supported client and scoped key.
- A buyer who wants assisted setup or a separately scoped customer-hosted pilot.
- For customer hosting, an operations owner who can manage backups, the disk and credentials.
- Certification: no SOC 2, ISO 27001, ISO 42001 or independent penetration test is held or in progress. Your own team, or one you commission, may test the source during the evaluation.
- Availability: there is no availability SLA, uptime percentage or service credits. A design-partner agreement is offered with first-response targets from a named person instead.
- Cloud topology: one replica with shared process, storage and traffic quotas; no high-availability commitment. Workspace processes share an OS user and host, so separation is not a sandbox or VM.
- Support hours: 09:00 to 17:30 Europe/London, Monday to Friday.
- Seat spend: caps on Copilot, ChatGPT or Cursor seats are preview. USD ceilings apply to the agents you route through the gateway.
- Workload: use this pilot for evaluation. Keep workloads whose outage could harm customers or regulated operations outside it.
Each of those is published in full, with the evidence behind it, on the trust page: what we do not claim.
What shapes a customer-hosted quote.
- Deployments
- One installation you control, plus the non-production copies the licence permits. A second line means a second production install.
- Simultaneously active agents
- Counted in the active state only, and enforced on control-plane writes, never on the request path, where a refusal would take a fleet down for a commercial reason.
- Entitled evidence modules
- Three, and only three: pull connectors holding a narrow read-scoped vendor credential you supply, push receivers for your own exporter or SIEM, and seat governance. Nothing that enforces is on the list.
- Subscription seats
- A second ceiling, for managed developer devices. The surface is preview and reports a technical blocker, so ask what it does today.
- Subscription term
- A hard wall in one direction: additions are refused and named, nothing already running is touched. A lapse refuses growth rather than causing an outage.
- The support relationship
- A design-partner agreement is a different commitment from a licence with no support attached, and is priced as one.
- Work that is not support
- Authoring or mapping a policy set. So is any certification milestone procurement requires: committable with a named date, and that date has a cost.
Customer-hosted licensing has separate capacity and operational terms. Read the trial scope before comparing them.
See the setupAgree support response targets.
These are proposed customer-hosted design-partner support terms, not a support commitment for the free cloud trial. They are first-response targets, never resolution times. You propose the severity, the higher one applies while it is disputed, and a signed agreement takes precedence.
Production agent traffic is stopped, or a control has failed in a way that lets ungoverned or unrecorded traffic through, with no workaround. This includes a deployment that serves traffic but has stopped recording it.
First response within 2 business hours, an update every 4, a same-day workaround where one exists, and a patch as soon as a fix is validated. No date is promised before diagnosis.
A major function is broken or materially degraded in production, but a workaround exists.
First response within 1 business day, daily updates, a workaround targeted within 5 business days, and a fix in the next release.
A function is broken or incorrect with a straightforward workaround, or the impact is confined to non-production.
First response within 3 business days, weekly updates, and a fix in the next release or a place in the backlog with the reason given.
A question, a documentation issue, a cosmetic defect or a feature request.
First response within 5 business days, then the backlog, with no commitment attached.
Hosting mode matters to availability.
There is no availability SLA, uptime percentage or service credits. Read how the single-node topology, fail-closed behaviour and available recovery evidence affect that position. Why there is no SLA.
What does the cloud trial cost?
The Token Observe cloud trial is free for 30 days from workspace creation, with no card and no automatic paid conversion. Paid cloud pricing is not published. Model calls, model-assisted search and Google BigQuery scans can incur charges with your providers. Customer-hosted licences and assisted work are quoted separately against a written scope.
What happens when the cloud trial ends?
Workspace changes, model calls, scheduled billing collection and paid-capable natural-language trace searches stop. Authorized retained GET reads and exports remain subject to the shared traffic quota. Exact first-party MCP reporting reads can continue with an active and currently authorized workspace key; unknown or upstream tools, batches and mutations remain refused. Nothing automatically converts to a paid subscription.
Does connecting billing also show historical conversations?
No. Billing APIs provide cost records, not historical prompts, answers, browser conversations or hidden chain of thought. Activity requires requests routed through the workspace gateway or another supported evidence source you explicitly configure. Missing telemetry is unknown coverage, not zero use.
Which provider credentials do we need?
OpenAI and Anthropic billing connections require their organization Admin API keys. xAI needs a separate management key and team ID. Google needs a configured Cloud Billing BigQuery export and the required reporting permissions. Model inference uses a separate key, while gateway and MCP clients use workspace-issued agent or seat keys. The portal rejects a stored billing secret reused for inference.
Can we still host Token Observe ourselves?
Yes, customer hosting remains available through assisted setup. The commercial source-available licence covers deployments, simultaneously active agents, optional support and separate professional services. It remains a template pending counsel. Customer-hosted data custody, recovery and licence-expiry behavior are separate from the hosted trial.
Is there an uptime SLA?
No availability SLA, uptime percentage or service credits are offered for the cloud pilot. It runs as one replica with shared resource and traffic limits. Customer-hosted design-partner support can state first-response targets in a signed agreement; those response targets do not establish service availability or cloud-trial support.
Can our security team test it before buying?
The software licence expressly permits inspection, fuzzing, penetration-testing and reverse-engineering of a deployment you control, including commissioned tests. Those permissions do not authorize testing the shared cloud service. Coordinate cloud security questions with the published security contact. Token Observe has no independent penetration test or SOC 2, ISO 27001 or ISO 42001 certification.
What must we agree before customer hosting?
Agree the deployment scope, active-agent capacity, support and fees in a signed agreement. The published licence remains a draft requiring counsel approval, completion of its placeholders and a liability-cap check against insurance cover. It states intended terms; your signed agreement governs.
Use the threat model, product limits and framework mappings to prepare your procurement review. Security and pilot limits · Framework mappings
Ask for the figure, and the scope it is quoted against.
Send the shape of the estate if you have it (deployments, agents active at once, providers), or send nothing but the question. Either way the quotation comes back against the definitions on this page, and the free 30-day trial does not wait for it.