The tools a developer's container may run. The container checks this list before every tool call and blocks anything not on it. Nothing listed means nothing is allowed. In audit mode an off-list tool runs but is flagged.

Latest tool calls a container refused (or flagged in audit mode) across the org. Admin only. Names, hosts and sizes only, never arguments or results.

The models your developers may use. The gateway checks this list on every call and refuses a model that is not on it, whether the developer named it directly or let IX pick by policy. In audit mode an off-list call runs but is flagged. Until you switch it on, every model is allowed.

Latest calls where the model was off the list. Admin only. Names only, never prompts or outputs.

Tools a simulated Agentic Workflow can draw on. Illustrative fixture data — not live connections yet. Costs shown are static per-call estimates, not real usage.

Companies providing observability for LLM/agent workloads elsewhere in the market. Informational reference only -- IX is not currently integrated with any of these.

Already run your agent's evals or traces through one of these? Route the same agent through IX and keep your existing tool for quality -- IX adds capability-weighted routing across providers, a real BitToken settlement ledger, and a compliance-grade audit trail (who spent what, on which provider, at what capability score, under what approved policy) alongside it, not instead of it.

Bring your own agent framework or SDK. Point your LangGraph/CrewAI/etc. model client at IX instead of a provider directly, and every call gets routed through IX's oracle and billed in BT — same real, per-call cost accounting as everything else in this platform. Readiness reflects what IX has actually verified, not assumed.

The coding agents your developers already use. Point them at IX and every call is routed through IX's oracle and billed in BT, with per-call cost, attribution to a ticket and project, and budgets, while the developer's terminal looks the same.

Your team's evaluation and tracing tools. IX complements them, it does not replace them: it can record how a run went and label eval traffic so it is kept out of retry and rate alerts.

Tell IX where your tickets live and every ticket id in the dashboard becomes a link to the real ticket. IX stores only the link pattern: it never signs in to these tools, never reads ticket titles or text, and never sends anything to them. Use {id} where the ticket goes ({num} for just the number).

For companies that buy their models through a cloud account they already have. The plan is for IX to sit in front of that account: your developers keep their tools, calls are forwarded using your own cloud credentials, and billing stays with the cloud provider while IX records cost and attribution. None of these is connectable yet.

Every real agent-matching decision from Project decomposition, with the full candidate set considered — not just the top-ranked one. Same role as the Oracle Market Log plays for LLM candidates. Real data from GET /agent-candidates — empty until a Project has been decomposed.

Viewing Project-level financials as Developer.

Set spending limits for your Project. Limits cascade downward — a Developer cannot exceed the Manager's department limit.

Convert fiat currency to BitToken. Card payments settle instantly via the notional payment rail; wire transfers are confirmed by our finance team once received.

Audit trail of every ledger movement. The settlement layer exposes why each provider was chosen — capability score, benchmark source, routing policy — not just the final BT debited. Enterprise-grade, append-only.

Commercial outcome of every execution. Think Bloomberg Terminal.
Click any row to see it in full in Call Detail.

Infrastructure health using OpenTelemetry conventions.
Two spans per execution: Routing (provider call) and Settlement (IX ledger write). BT Cost is the real spend for the execution as a whole, shown on its Routing span.

How each piece of work unfolded: calls in order, what each cost, where it looped, was refused or sat idle. Built only from ledger metadata: no prompts, code or output are held. An administrator sees the organisation, a manager their people, a developer their own. Opening someone else's session is recorded.

One call, in full: what it was, which model and route, what it cost and why, what was flagged, where it sits in its session, and what that leaves of your budget. Every value is read from the ledger; anything IX does not have shows "—".

How far each model can be trusted to deliver a task on budget. The estimate is the one written before the first call; the cost is what was really charged. Built from every dev run recorded so far. Rows with fewer than 5 runs show a dash rather than a rate off too little data.

Real, measured results from the Isomorphic Labs / Lewis Silkin outcome-pricing pilot. Every figure below is computed directly from actually-logged runs — nothing here is estimated or backfilled.

Real, first-party success/failure data from every live provider attempt IX has made — across every execution, not just yours. Every row is a genuine outcome (adapter is not None): no simulated calls counted.