Your organization's AI use is opaque.
ixway clears the fog.
ixway in 40 seconds
You know what you spent on AI last month. Do you know where it went? Requests from harnesses and SDKs flow toward the models and disappear.
ixway sits between them and every request passes through it. Intelligent routing: routing you can see and set, the right model for each job. Insight extraction: know how your workforce uses AI and what it actually produced, with spend by type of work (feature work, debugging, customer support, research) and the pull requests and deploys it is linked to. Local and batch inference: your own hardware first, batch work in cheaper windows. Efficiency findings: know where the waste is. One finding, a volatile header breaking the cache prefix every turn, is sent to the harness that caused it, and that harness's tokens per session fall by 38% once it is applied. ixway teaches your harnesses to be more efficient. The figures are illustrative.
Reduce your spend. Stretch your tokens. Transparency and efficiency for your workforce AI usage. ixway.ai.
01Observability
Know what your agents did, what it cost, and why.
ixway traces every session end to end. Each agent, each subagent it spawned, each model call, each tool call with what it actually did.
Goal · inferred from session sess_4e1b
Add idempotency keys to the payments API and backfill 1.2M existing orders
- $3.52 est. avoidableReplay billed above cache rate361,000 tokens re-sent uncached at $10/M after the x-request-id header changed the prefix 61 times; as cache reads they cost $0.25/M
- $1.38 est. avoidableModel and effort mismatch9 of 31 Fable turns were one-line edits; Sonnet 5.5 passed the same turns on replay
- $0.36 est. avoidableUnused tool definitions13 of 31 definitions never called (ask_user, skill and 11 smaller) · 263,000 of 641,390 definition tokens re-sent with every request
- writing to you 34m 37s · 44%
- composing tool calls 16m 11s · 21%
- waiting on the model 13m 5s · 17%
- running tools 10m 9s · 13%
- delegating 2m 53s · 4%
- reasoning 59.0s · 1%
Hover to preview · click a tool to filter the tape
- 21:26:00agent → bashbash · 34.2s · 295 chars backSend the same webhook retry to staging twice and confirm the second reuses the stored response.$0.12so far
- Turn
- 39 · claude-fable-5-1 · 21:25:56 → now, in progress
- Tokens in
- 101,300 = 1,100 uncached · 4,200 cache write · 96,000 cache read
- Tokens out
- 600
- Turn cost so far
- $0.1175 = (1,100×$10 + 4,200×$12.50 + 96,000×$0.25 + 600×$50) / 1M
- This call
- bash · curl staging /webhooks/stripe twice · round trip 34.2s · 295 chars back (≈74 tokens)
- Came back
- HTTP/1.1 200 both times · idempotent-replayed: true · one charge, ch_3Qx…a41
- Turn so far
- maya asked about retries; the agent is proving a retried webhook cannot double-charge.
- 21:25:59agentreasoned · 131 tokens$0.12so far
- Turn
- 39 · claude-fable-5-1 · 21:25:56 → now, in progress
- Tokens in
- 101,300 = 1,100 uncached · 4,200 cache write · 96,000 cache read
- Tokens out
- 600
- Turn cost so far
- $0.1175 = (1,100×$10 + 4,200×$12.50 + 96,000×$0.25 + 600×$50) / 1M
- This step
- 131 reasoning tokens, billed as output inside the turn’s 600
- 21:25:56you → agentsent a messageAsked whether webhook retries reuse the original idempotency key.—
- Turn
- 39 starts here · claude-fable-5-1 (session pinned)
- Message
- 41 tokens, after 11m 21s of idle
- Cost
- a message costs nothing until the model reads it; it is part of turn 39’s input
- 21:14:32agent → editedit · 1.2s · 312 chars backScope the key lookup in idempotency.ex to merchant_id so two merchants can reuse a key.$0.89
- Turn
- 38 · claude-fable-5-1 · 21:09:15 → 21:14:34
- Tokens in
- 276,200 = 61,400 uncached · 9,800 cache write · 205,000 cache read
- Tokens out
- 2,100
- Turn cost
- $0.8928 = (61,400×$10 + 9,800×$12.50 + 205,000×$0.25 + 2,100×$50) / 1M
- This call
- edit lib/payments/idempotency.ex · round trip 1.2s · 312 chars back (+4 −1)
- Turn summary
- Found where the charge handler must check the key before capture, scoped lookups to merchant_id.
- 21:14:22agent → readread · 143ms · 6.9k chars backRead payments/handler.ex to find where the key must be checked before capture.$0.89
- Turn
- 38 · claude-fable-5-1 · 21:09:15 → 21:14:34
- Tokens in
- 276,200 = 61,400 uncached · 9,800 cache write · 205,000 cache read
- Tokens out
- 2,100
- Turn cost
- $0.8928 = (61,400×$10 + 9,800×$12.50 + 205,000×$0.25 + 2,100×$50) / 1M
- This call
- read payments/handler.ex · round trip 143ms · 6.9k chars back (≈1.8k tokens)
- Came back
- handle_charge/2 calls Charge.capture/1 before any key lookup
- F-0398
- prefix changed 10× by the x-request-id header · billed 61,000 uncached · $0.61 · as cache reads it would have been $0.015 (61,000×$0.25/1M) · est. avoidable $0.59
- 21:09:19agent → grepgrep · 0.9s · 2.1k chars backFind every call site that builds an idempotency key before touching the handler.$0.89
- Turn
- 38 · claude-fable-5-1 · 21:09:15 → 21:14:34
- Tokens in
- 276,200 = 61,400 uncached · 9,800 cache write · 205,000 cache read
- Tokens out
- 2,100
- Turn cost
- $0.8928 = (61,400×$10 + 9,800×$12.50 + 205,000×$0.25 + 2,100×$50) / 1M
- This call
- grep -rn idempotency_key lib/ · round trip 0.9s · 2.1k chars back (≈560 tokens)
- Came back
- 14 matches across 6 files; 3 in handler.ex
- 20:01:50you → agentsent a messageRun the suite again and fix whatever the backfill broke.—
- Turn
- 37 starts here · claude-fable-5-1 (session pinned)
- Message
- 19 tokens
- Cost
- a message costs nothing until the model reads it
- 19:59:24agent → writewrite · 0.6s · okWrite the migration that adds idempotency_keys and backfills existing orders in batches of 10k.$0.05
- Turn
- 35 · claude-fable-5-1 · 19:59:17 → 20:00:35
- Tokens in
- 33,573 = 169 uncached · 1,899 cache write · 31,505 cache read
- Tokens out
- 425
- Turn cost
- $0.0546 = (169×$10 + 1,899×$12.50 + 31,505×$0.25 + 425×$50) / 1M
- This call
- write priv/repo/migrations/20261008_add_idempotency_keys.exs · 2.3k tokens out · round trip 0.6s
- Turn summary
- Migration written with a batched backfill and a unique index on (merchant_id, key).
- 19:59:18agentreasoned · 904 tokens$0.05
- Turn
- 35 · claude-fable-5-1 · 19:59:17 → 20:00:35
- Tokens in
- 33,573 = 169 uncached · 1,899 cache write · 31,505 cache read
- Tokens out
- 425
- Turn cost
- $0.0546 = (169×$10 + 1,899×$12.50 + 31,505×$0.25 + 425×$50) / 1M
- This step
- 904 reasoning tokens, billed as output inside the turn
- 19:57:31you → agentsent a messageBackfill has to run in batches; 1.2M rows in one transaction will lock the table.—
- Turn
- 34 starts here · claude-fable-5-1 (session pinned)
- Message
- 27 tokens, after 2m 10s of idle
- 19:56:57agent → todotodo · 40ms · 6 itemsTrack the remaining work: migration, backfill, handler change, tests, PR.$0.67
- Turn
- 33 · claude-fable-5-1 · 19:56:45 → 19:57:30
- Tokens in
- 204,586 = 30,194 uncached · 10,292 cache write · 164,100 cache read
- Tokens out
- 4,016
- Turn cost
- $0.6724 = (30,194×$10 + 10,292×$12.50 + 164,100×$0.25 + 4,016×$50) / 1M
- This call
- todo write · 6 items · round trip 40ms
- 19:56:48agent → readread · 98ms · 4.1k chars backRead lib/payments/charge.ex to see how captures are committed today.$0.67
- Turn
- 33 · claude-fable-5-1 · 19:56:45 → 19:57:30
- Tokens in
- 204,586 = 30,194 uncached · 10,292 cache write · 164,100 cache read
- Tokens out
- 4,016
- Turn cost
- $0.6724 = (30,194×$10 + 10,292×$12.50 + 164,100×$0.25 + 4,016×$50) / 1M
- This call
- read lib/payments/charge.ex · round trip 98ms · 4.1k chars back (≈1.1k tokens)
- Came back
- Charge.capture/1 commits inside the request; no key is recorded
- 19:56:46agent → globglob · 31ms · 9 paths backList the payments modules before reading any of them.$0.67
- Turn
- 33 · claude-fable-5-1 · 19:56:45 → 19:57:30
- Tokens in
- 204,586 = 30,194 uncached · 10,292 cache write · 164,100 cache read
- Tokens out
- 4,016
- Turn cost
- $0.6724 = (30,194×$10 + 10,292×$12.50 + 164,100×$0.25 + 4,016×$50) / 1M
- This call
- glob lib/payments/**/*.ex · round trip 31ms · 9 paths back (≈120 tokens)
- 17:55:09test-runner → bashbash · 41.0s · 38 lines backRun the payments suite after the fix; all 12 tests pass.$0.00local
- Turn
- 28 · test-runner on local qwen3-coder-480b · 17:55:03 → 17:55:13
- Tokens in
- 24,832 · served locally, nothing billed
- Cost
- $0.00 measured · cloud equiv $0.0055 at $0.22/M
- This call
- bash mix test test/payments · round trip 41.0s · 38 lines back (≈310 tokens)
- Came back
- 12 tests, 0 failures
- 17:54:57test-runner → editedit · 0.8s · 212 chars backReturn the stored response when a key is seen twice instead of raising.$0.00local
- Turn
- 27 · test-runner on local qwen3-coder-480b · 17:54:54 → 17:55:03
- Tokens in
- 21,565 · served locally, nothing billed
- Cost
- $0.00 measured · cloud equiv $0.0047 at $0.22/M
- This call
- edit lib/payments/handler.ex · round trip 0.8s · +14 −3
- Came back
- ok
- 17:54:46test-runner → readread · 121ms · 3.4k chars backRead the failing test to see what it expects on a duplicate key.$0.00local
- Turn
- 26 · test-runner on local qwen3-coder-480b · 17:54:44 → 17:54:54
- Tokens in
- 23,699 · served locally, nothing billed
- Cost
- $0.00 measured · cloud equiv $0.0052 at $0.22/M
- This call
- read test/payments/idempotency_test.exs · round trip 121ms · 3.4k chars back (≈900 tokens)
- Came back
- expects {:ok, stored} on the second call, got {:error, :duplicate_key}
- 17:54:39test-runner → bashbash · 39.4s · 38 lines backRun the payments suite; one idempotency test fails.$0.00local
- Turn
- 25 · test-runner on local qwen3-coder-480b · 17:54:34 → 17:54:44
- Tokens in
- 44,345 · served locally, nothing billed
- Cost
- $0.00 measured · cloud equiv $0.0098 at $0.22/M
- This call
- bash mix test test/payments · round trip 39.4s · 38 lines back (≈310 tokens)
- Came back
- 12 tests, 1 failure: idempotency_test.exs:41
- 17:53:08agent → delegatedelegate · spawned test-runnerRun the payments suite and fix the failing idempotency test.$0.75
- Turn
- 22 · claude-fable-5-1 · 17:52:59 → 17:56:36
- Tokens in
- 285,269 = 36,290 uncached · 10,980 cache write · 237,999 cache read
- Tokens out
- 3,743
- Turn cost
- $0.7468 = (36,290×$10 + 10,980×$12.50 + 237,999×$0.25 + 3,743×$50) / 1M
- This call
- delegate test-runner · brief 1.2k tokens · routed by ixway to local qwen3-coder-480b (local_first_meets_floor)
- Came back
- 4 local turns, $0.00 · summary 410 tokens returned to the main agent
Inside system prompt · priced at this bucket’s billed mix $0.444/M (20,000 uncached · 9,000 write · 1,175,813 read · 160,000 local)
- harness instructions78%1,064,554$0.473
- CLAUDE.md files15%204,722$0.091
- environment7%95,537$0.042
- ask_user26.0%
- delegate14.8%
- grep12.2%
- bash11.8%
- todo8.4%
- read6.9%
- glob6.2%
- edit5.3%
- skill5.2%
- write3.3%
Inside tool definitions · priced at this bucket’s billed mix $1.355/M (30,000 uncached · 34,000 write · 577,390 read · 0 local)
- ask_user1 def · called ×0166,575$0.226never called
- delegate3 defs · called ×394,887$0.129delegate_status · never calleddelegate_cancel · never called
- grep3 defs · called ×2178,435$0.106grep_count · never called
- bash4 defs · called ×13275,386$0.102bash_kill · never calledbash_background · never called
- todo4 defs · called ×2053,587$0.073todo_clear · never called
- read4 defs · called ×12444,044$0.060read_notebook · never calledread_image · never called
- glob3 defs · called ×1239,577$0.054
- edit5 defs · called ×5833,907$0.046notebook_edit · never calledapply_patch · never called
- skill1 def · called ×033,563$0.045never called
- write3 defs · called ×621,429$0.029write_notebook · never called
13 of 31 definitions never called · 263,000 tokens · $0.36 re-sent with every request (F-0412)
Inside tool results · priced at this bucket’s billed mix $1.328/M (70,000 uncached · 195,000 write · 2,263,972 read · 260,000 local)
- file reads · read60%1,673,383$2.222
- command output · bash29%808,802$1.074
- search · grep, glob8%223,117$0.296
- other · edit, delegate, todo, write3%83,670$0.111
Inside user · priced at this bucket’s billed mix $7.760/M (6,000 uncached · 12,000 write · 9,363 read · 0 local)
- maya’s 11 messages64%17,512$0.136
- harness reminders36%9,851$0.076
Inside assistant text · priced at this bucket’s billed mix $2.422/M (9,000 uncached · 20,000 write · 90,732 read · 30,000 local)
- replies to maya72%107,807$0.261
- progress notes28%41,925$0.102
Inside tool calls · priced at this bucket’s billed mix $1.911/M (20,817 uncached · 30,000 write · 235,065 read · 50,000 local)
- edit diffs46%154,506$0.295
- bash commands38%127,635$0.244
- other calls16%53,741$0.103
turn 1 · 1,362,598 · 25.7% · $1.75 · Read the payments handlers and planned the idempotency-key migrationnot in thread 793,730 · test-runner ×16 500,000 · $0.00 local (cloud equiv $0.12)
billed input 5,532,335 · analysed 5,308,152 · $11.34 measured · baseline $58.39 all-cloud, no cache
This session is PR #482, one tile in platform-eng's month.
See what leaders see ↗02Routing
Routing you understand and control.
Not an auto button. You pick the strategy, set a policy per team or agent cluster, and every decision is written down: the candidates, the choice, the reason.
profile eng/coding · strategy: capability floor, then lowest cost · shadow: off · session pinned for cache continuity
“Plan the idempotency-key refactor across the payments handlers”
- intent
- refactor
- complexity
- high
- files
- 6
- context
- 38k tokens
- priority
- normal
- max cost
- $0.60
- claude-fable-5-1chosen anthropic · cloud · est $0.1934 · p50 9.1 s meets capability floor The only candidate at or above the floor for a high-complexity refactor. $0.1934 fits the $0.60 cap; cache reads keep it under half the $0.4750 uncached cost.
- claude-opus-5-5 anthropic · cloud · est $0.0803 below floor for this task $0.1131 cheaper, but it scores under the floor for multi-file refactors. Cost only breaks ties above the floor.
- claude-sonnet-5-5 anthropic · cloud · est $0.0401 below floor for this task Cheaper again, and further under the floor for a change that spans six handlers.
- qwen3-coder-480b vllm · rack-a · local · $0.00 · cloud equiv. $0.0103 below floor for this task Free on the local rack, but under the floor here. It clears the floor for simpler work: see B.
x-ixway-resolution-reason: capability_floor_cheapest
baseline, all on fable, no cache: $0.4750
“Run the payments suite and fix the failing idempotency test”
- intent
- test-fix
- complexity
- low
- files
- 2
- context
- 9k tokens
- priority
- normal
- max cost
- $0.60
- qwen3-coder-480bchosen vllm · rack-a · local · $0.00 · p50 2.8 s meets floor · local first · cloud equiv. $0.0026 Clears the floor for a low-complexity test fix, so the local rack takes it: $0.00 billed, $0.0026 at the hosted list price.
- claude-haiku-5-5 anthropic · cloud · est $0.0012 · p50 2.1 s eligible · local scored higher The cheapest and fastest cloud option, but it scored below qwen3-coder-480b on test fixes, and local bills nothing.
- claude-sonnet-5-5 anthropic · cloud · est $0.0244 eligible · unnecessary for this task Clears the floor, but nothing here needs it. $0.0244 for the same outcome.
- claude-fable-5-1 anthropic · cloud · est $0.1220 eligible · unnecessary for this task The most expensive way to fix one test in two files: $0.1220 for what the local rack does at $0.00.
x-ixway-resolution-reason: local_first_meets_floor
list prices · fable 5.1 $10/$50 · opus 5.5 $4/$20 · sonnet 5.5 $2/$10 · haiku 5.5 $0.10/$0.50 · qwen3-coder-480b $0.22/$1.00 per M tokens in/out
- 2.1Policies per team. Profiles bind to an organization, a unit, a project, or a single agent.
- 2.2Strategies you choose. Lowest cost, lowest latency, capability-gated, peak and off-peak economics, classifier, escalate on a weak answer, or a judge trained on your own traffic.
- 2.3Shadow first. A new policy records what it would have done until you switch it on. Sessions stay pinned so caches keep hitting.
- 2.4Explained in the response. Resolved model, backend, reason, and decision id come back as headers. Replay any policy offline before you ship it.
02.1Local inference
Get everything out of the hardware you own.
When local can handle the request, local gets it. ixway balances your fleet, keeps priority work inside its SLA, and diverts to cloud only under pressure.
p=high ×3
dispatched first
→ haiku 5.5 · $0.0242
local 257 req · $0.00 (cloud equiv. $0.7351) · cloud 97 req · $3.8048
- 2.5Real engine signals. vLLM, SGLang, llama.cpp, TensorRT-LLM, Ollama. Queue depth and capacity read straight from the engine.
- 2.6Priority dispatch. Tag work with a priority and a latency SLO. The important traffic keeps its place in the queue.
- 2.7Honest economics. Every local request shows its cloud-equivalent cost. You finally know what the rack is worth.
03Efficiency insights
Find the waste. Hand the fix to the harness.
ixway produces findings the way a scanner produces vulnerabilities: specific, evidence-backed, priced. Your harnesses read them and get leaner.
- 3.1Findings, not advice. Each one links to the requests, tool calls, and sessions where it happened.
- 3.2Three honest numbers. Measured spend, estimated avoidable spend, and verified savings after the fix. Never one pretending to be another.
- 3.3A loop, not a report. Quotes before a request, receipts after, compaction advice when context bloats, findings back to the harness.
- 3.4Verified after. Accept a fix, re-run comparable work, and see whether the saving was real.
Redefine tokenmaxxing. Maximize value per token.
04System of record
A system of record your harnesses learn from.
The ledger is the product. Every view above reads from it. So do your harnesses.
- 4.1
Tamper-evident ledger
Every request is a hash-chained row. Verify the chain any time. Reconcile it to provider invoices at period close.
- 4.2
Ask it anything
A query builder with CSV export, and an assistant that answers with cited queries, routed through your own gateway.
- 4.3
Outputs, not requests
Sessions are classified by activity and outcome and linked to the commits, pull requests, and deployments they produced.
- 4.4
Application traces, joined
Send OpenTelemetry traces from your own agents. ixway joins them to the ledger, so a workflow run shows its cost per outcome.
- 4.5
Budgets that bite
Daily, weekly, and monthly budgets per key. Rate limits, priority queues, and batch APIs for work that can wait.
- 4.6
Privacy by default
Capture is optional. Individual reporting hides behind minimum group sizes. Content never leaves your deployment unless you say so.
05How it works
Point your inference at ixway. Keep everything else.
ixway is a gateway. Harnesses and SDKs talk to it exactly as they talk to a provider. It speaks the Anthropic and OpenAI dialects natively and relays them byte for byte.
harnesses & sdks
- Lemieux · lmx
- Claude Code
- Codex
- Anthropic SDK
- OpenAI SDK
- LangChain · LangGraph
- OpenAI Agents
- OpenInference
gateway
- passthrough
- Your Claude or ChatGPT login goes straight through. ixway holds no credential.
- gateway keys
- Provider keys live in ixway. People and agents get virtual keys with budgets and limits.
- telemetry
- Your own agents send traces on a separate credential. Nothing can spend with it.
models
- Anthropic
- OpenAI
- 30+ cloud providers
- vLLM
- SGLang
- llama.cpp
- TensorRT-LLM
- Ollama
local engines