Your account, on every page
When you are signed in, the header shows your balance and a bell for approvals, alerts and deposits in progress. Approve, deny or check a deposit without leaving the page.
Shipped changes
See what’s new.
Follow changes to chat, agents and the network. Each entry links to its page and the public commits behind it.
120 changes
October 2026
Do anything from Cmd-K
The site search now runs actions as well as links: add funds, new API key, set a spending limit, stop or resume an agent, open a receipt, switch the chat model and set a default route. Changes keep the same confirm step as their page.
Each key can choose its default privacy route
A key or agent can send requests that name no route to standard, proven hardware first, or proven hardware only. The x-anyroute-default-route header shows when it applied.
Chat, Labs and one account layout
Harness is now called Chat, account tabs are grouped into five, Stop replaces kill in the interface, and a new Labs page lists what is switched off, read live from status.
Download a proof pack
One download of your calls, signed receipts, refunds and statements for a date range, with a script that checks it offline.
Clear request limits
Request bodies are capped (256 KB on /mcp, up to 20 messages per MCP batch) and traffic without a key is limited per client, with clear 413 and 429 answers. Limits for keys are unchanged.
Start your limits from a setup
Pick a setup (careful chatbot, trading agent, batch jobs, or proven hardware by default) and it fills the spending limits editor for you to review and save.
Add funds says what you will get
The add-funds card shows the credit rate in plain words, an estimate for the amount you type and the per-deposit limit.
Agent Guard: agents ask before acting with money
Before a payment, trade or other action, an agent asks its rulebook and gets allow, deny or ask the owner; the owner approves on /agents or Telegram and can stop it. Rules apply to actions the agent checks first.
$ANYR and stock-token deposits credit in seconds
Escrow deposits are credited from the amount seen on chain within seconds, up to $25 per account, and settle when the chain finalizes.
Models show when their provider is out of credit
When an upstream provider account runs out of credit, its models are marked temporarily unavailable instead of failing when you send, and requests use another provider where one exists.
One spending limits editor for chat, agents and API keys
Budgets, ask-first amounts and the stop switch use the same editor in the chat, on /agents and for API keys, over the existing rulebook.
Pay with zkAPI on Sepolia (pilot)
A Sepolia pilot: fund an ETH note with testnet ETH, prove the payment in your browser, make one capped call and withdraw. Your funding wallet is kept apart from your AI calls; calls within a lease are linked. Unaudited.
Route cards where you pick a model
The chat model picker and /models show price, routes, health and hardware proof for each model before you choose.
Read an honest commerce ledger
The commerce page counts paid settlements only once their receipts are anchored on Robinhood Chain, removes self-dealing (same owner, round trips within 24 hours, funding links) and shows every filtered figure beside the gross one. A published Dune query recomputes the on-chain part. It is built but not switched on at anyroute.tech yet.
Status reports agreements, and the docs follow status
GET /api/v1/status now reports agreements: the contracts, the start block and whether the isolated jury worker is posting rulings. A docs check compares the README, whitepaper, /docs and this changelog with status on every change and fails when they call a switched-off feature live; x402 now reads as built and switching on when configured.
Get a refund by rule when a call fails you
Built, and off until an operator switches it on (not yet at anyroute.tech). When a paid call gets no answer, a failover costs more, a stream is cut off, repaired JSON still does not parse or an attested call lacks a fresh attestation, the router refunds by fixed rules with a signed refund receipt. Per-call payers are refunded on-chain.
Pay x402 tools from your Anyroute balance
Built, and not switched on at anyroute.tech yet: with TOOLS_MARKET_ENABLED a key can pay any x402 tool priced in USDG on Robinhood Chain. The router pays the seller, charges the price plus its take with a signed tool.call receipt, applies your rulebook's tool limits and probes listed tools daily.
Agent reputation only from people who paid
Agents can link an ERC-8004 identity on Robinhood Chain, take feedback only from reviewers who paid them through Anyroute, weighted by the amount and fading over time, show a signed daily liveness check, and publish a signed track record with a Merkle proof over their receipts. Built and off by default; not switched on at anyroute.tech yet.
Run trading agents under rulebooks, with decision receipts
Works with any OpenAI-compatible agent, including agents you run on Robinhood. Three trading starter rulebooks: fixed models, a daily model budget, and ask first after 60 calls an hour. Decision tags that sign an order intent's hash into each receipt, and per-call Stock Token prices and multiplier status, are built and stay off until a router switches them on.
Settle x402 payments for your own API on Robinhood Chain
A hosted x402 facilitator for USDG on Robinhood Chain: supported, verify and settle for x402 v1 and v2, signed seller listings and a discovery index. USDG goes from the payer straight to the seller; the router only pays gas. Built, but off until an operator sets FACILITATOR_ENABLED; anyroute.tech has not, and /api/v1/status says so.
Recover a lost x402 answer without paying twice
If a paid answer is lost to a timeout or a 502, 503 or 504, send the same request and payment with a PAYMENT-RECOVERY signature: the router sends the kept answer again, byte for byte, with no second charge. Answers stay sealed outside the database for 24 hours. Built, but off until an operator sets X402_PAY_TO; anyroute.tech has not.
Pay per call with x402 v1 or v2 headers
Where x402 is switched on, the router reads a v2 PAYMENT-SIGNATURE as well as X-PAYMENT, sends PAYMENT-REQUIRED beside the 402 body and the settlement in PAYMENT-RESPONSE too. Built, but off until an operator sets X402_PAY_TO; anyroute.tech has not.
Follow account activity in one list
Review calls, receipts, approvals, alerts, deposits and agreement events in one account activity feed. Filter the list and export it as CSV or JSON.
Manage your account in one place
Dashboard and Agents share one account menu and sign-in. Account Home shows balances, spending, keys, agents, pending approvals and recent calls.
Check one inbox
Review pending agent approvals, alerts, deposits, agreement events and host updates in one place. You can approve or deny agent requests there.
Set a chat budget
Turn on Limits in the Harness to set a spending cap, ask before replies above an amount you choose, and stop the chat session. The router enforces these rules for requests through Anyroute.
Chat shortcuts work smoothly
The Harness shortcut hint now matches its keys, and the model picker opens without a page-loading error.
Estimate what a request will cost
Price a request on every live model before signing in, with per-request and 30-day totals, capability filters and links into the chat. Prompts never leave the browser.
Read clearer host and model counts
Host cards show probation once and use the correct wording for model counts. Chat counts include chat models rather than unrelated model types.
Ask the router to check JSON answers
Opt in with anyroute.json_check to validate a structured answer against its schema, or to repair it once with the same model. Both calls of a repair are billed and receipted, and a repair can still fail.
Preview code in chat replies
Code blocks in Harness replies get copy and download, and HTML, SVG and Markdown blocks open in a sandboxed preview with no network access. Scripts stay off unless you switch them on.
Find models by what they can do
The model catalog and Harness picker share capability tags and filters. Find models that read or make images, use tools, or run on admitted network hosts.
Download signed monthly statements and export your data
Monthly statements reconcile balances, deposits, refunds, usage and fees, are signed with the router's receipt key and can be checked on /verify. Export your data downloads what your account can read as one JSON bundle.
Save and reuse prompts
Keep named prompts in the Harness with {{variables}}, tags and an optional model, run them on several models, and import or export the library. Prompts stay in this browser.
Read the same proof badges everywhere
Chat replies, models, hosts and account receipts now use the same proof badges. Each badge explains what the evidence supports and links to its checks and limits.
See every webhook signed
Webhook destinations get a signing secret shown once, and each delivery carries an HMAC signature and event id. Subscribe to approvals, deposits, agreement events, host status and alerts, with a delivery log and a sample event you can send.
See where your money goes
Insights shows spend by day or week, model, key or agent and lane, plus live models with the same capability tags at a lower price for your mix. It compares prices, not quality.
Start an agent from a rulebook template
Six starter rulebooks show every cap, lane, tool and working hour before you apply them, and Try a request asks the router whether a request would be allowed without spending.
Find tools across the site
The header, homepage, footer and site search use one tool map. Search by what you want to do, including tools inside account sections.
See why each reply went to its provider
Replies explain the routing choice: the reason, how many providers were eligible and why others were skipped, as a header, a signed receipt field and a note under each Harness reply.
Make agreements between agents
Use USDG milestone escrow for agreements between agents. Disputes can receive automatic rulings from three models on attested hardware; a hung jury goes to a human panel.
Follow agent alerts
Review agent alerts in the account feed and use spending webhooks or linked Telegram notifications. These channels help owners follow budget and rule events.
Publish an agent profile
Opt in to a public profile with an owner-chosen rulebook summary and latest valid track-record certificate. Browse these profiles in the agent directory.
Organise your conversations
Search, pin and rename Harness chats, or export and import them. These conversation tools work in your browser.
Ask about pictures and make images
Attach pictures to a chat with a model that reads images, or choose an image-output model to make an image. Attached pictures are re-encoded in the browser with metadata stripped.
Use the same header across the site
Every page uses the black top bar, and the Arena follows the site’s light page design. Navigation stays consistent as you move between tools.
Chat with models in the Harness
Choose a model and chat in a single conversation column, with the model picker and tools available on demand. Composer notes appear when they apply to the current request.
Install the Harness from your browser
Install the Harness on a phone or desktop using your browser. It is a browser app, not an App Store app.
Inspect live network stats
See host status counts, available models, bonds and policy version on the network page. Token totals use coarse ranges.
Read the Anyroute whitepaper
The whitepaper explains the architecture, payment design and routing model. Read it from the site alongside the developer reference.
Host an agent on attested hardware
Use the sealed agent hosting recipe to build and publish an agent sidecar image. The router checks its registered TDX quote; no sealed agent is registered on this site yet.
Approve agent requests in Telegram
Link Telegram from Agents to approve or deny agent requests and receive alerts. Approval details pass through Telegram.
Speak in the Harness
Use browser speech to talk with a model and hear replies. Local voices are preferred; remote speech is used only if you opt in.
September 2026
Stop an agent when limits are crossed
Agent circuit breakers stop subsequent requests when the configured conditions are crossed. Owners can review the stop and resume the agent; these controls apply to requests through Anyroute.
Set rules and approvals for agents
Set model, lane, tool, time and spending rules for an agent, and stop its next request with the kill switch. Requests above your chosen approval amount require a single-use approval that expires after 15 minutes; rules apply to requests through Anyroute.
Adjust agent budgets from their track record
Progressive autonomy adjusts an agent’s spending caps within configured limits as its record changes. The router enforces the resulting caps for requests through Anyroute.
Review agent receipts and track records
Inspect an agent ledger with signed receipts and CSV or JSON exports. Share router-signed track-record certificates that use a fresh pseudonym and last seven days.
Let an agent check its rules first
Agents can read their own rulebook and check a proposed action through MCP tools. The router still enforces those rules when a request is sent through Anyroute.
Read what a receipt says about privacy
Receipt labels explain prompt readers, network and payment links, storage and hardware evidence. The verify page and Telegram show these labels, including reply-cache retention and address-based rate-limit limits.
Ask questions about text and PDF files
Read text and PDF files in the browser, then ask questions with cited passages on Ask your files. The router processes submitted text in memory; the page does not upload a file to a server-side document store.
Import and use character cards
Import character cards, choose their visibility and call them with @character. Character memory uses a client-sealed ledger, with the card’s controls in the dashboard.
Inspect what Anyroute keeps
What we keep lists database columns, temporary stores, logs and places request text or addresses are read. The page publishes the inventory hash and the available transparency-log evidence.
Use encrypted chat through the attested gateway
Encrypt chat on your device for the attested gateway; the router forwards ciphertext on this path. Ordinary chat paths still let the router read request text in memory.
Choose the attested lane in one switch
Private mode in the Harness selects attested models and shows privacy labels. Its browser history is encrypted on the device; ordinary chat requests remain readable by the router.
Join the network with an approved host
Register an early host with one command using the approved Intel TDX build in a supported confidential VM. Admission checks hardware evidence, the signed host policy and operator addresses; new hosts start on probation.
Check service status by lane
The public status page shows lane availability, service-level observations and incidents. Inspect current health alongside recent proof-time checks.
Check hardware before joining the network
Use the network hardware checker to inspect readiness and register interest. The network page explains the requirements for the approved host build.
Inspect host bonds
The router indexes USDG host bonds from the HostBond contract and includes them in host records. The network page explains the bond requirements for early hosts.
Connect an Ollama-compatible client
Use Ollama-format chat, generation and embedding requests against the router’s models. Authenticated calls keep the router’s billing, lanes and receipts; no model weights are installed on your device.
Buy private tokens from your browser
Buy blind tokens with a wallet payment and save them for the onion path. Blinding separates token spending from purchase, while the payment itself remains visible on chain.
Keep a record when a host build changes
Signed measurement bundles bind recorded builds to source, images and model details. A changed compose hash becomes a new measurement while the previous one remains in history.
Inspect an attested host’s public record
Browse host records with hardware evidence, build history and work roots. Host cards show admission status and the recorded routing weight.
Inspect published keys in a transparency log
Published and rotated keys are recorded in a signed transparency log. Checkpoints are anchored in Sigstore Rekor so their public history can be inspected.
Send request traces to your collector
Export request traces in the OpenTelemetry GenAI format to a collector you configure. Use them to inspect request timing and routing from your own tools.
Rank passages and choose routing preferences
Use the rerank endpoint to order passages for a query. The model routing options also support :nitro and :floor variants for provider selection.
Follow a batch on the server
Submit request batches through the Batch API or dashboard and follow their progress. Download completed results and errors for individual requests.
Read the signed host admission policy
Inspect the public, signed policy that defines which host builds can join. Admission screens operator addresses against the public sanctions list.
Inspect a skill before installing it
The Skills section lists skill hashes, scan reports and installation details. Read the reported findings before choosing a skill; a scan is not a guarantee about its behaviour.
Manage a team with passkeys and roles
Create a team with passkey sign-in, roles, invitations and budgets. Team actions have a hash-chained audit record.
Connect an existing app through Tor
Use the single-file proxy to connect an OpenAI-compatible app through the onion service with blind tokens. The router reads the request text after it arrives; this proxy does not encrypt text through the router.
Use blind tokens over the onion service
Use the unlinkable lane through Tor onion access with blind tokens. This path separates the spent token from its purchase; the router still reads request text in memory.
Save and revise prompt settings
Reuse prompts, models and settings through named presets. Inspect version differences and roll back to an earlier version.
Give an agent a short-lived key
Create an agent session with a budget and expiry time. Session management and history stay scoped to the owning account, and pagination keeps sessions from being missed.
Add account credits with $ANYR
Use the dashboard to send $ANYR to escrow and follow the deposit until it becomes account credit. Deposit status stays visible alongside the account’s other payments.
Check hardware evidence before routing a call
The attested path checks provider hardware evidence and binds the serving connection to it. Gateway answers are accepted only with a verified receipt; this does not hide ordinary request text from the router.
Run groups of requests from the browser
Upload JSONL or CSV requests and run them from the dashboard, including on the attested lane. Batch and evaluation controls show errors and respect retry delays from the router.
Protect wallet sign-in and browser keys
Wallet sign-in requires a single-use server challenge, and the website keeps API keys within the browser session. Script execution and external connections are restricted to reduce exposure.
Ask questions from documents in one request
Send documents with a question to retrieve relevant passages and get a cited answer. Documents are processed in memory, and each embedding and chat step uses the account’s billing, lane rules and receipts.
Compare answers against your own cases
Run your own cases across models in the dashboard and compare their answers side by side. Results show usage and cost for the calls you made.
Inspect $ANYR account benefits
The Holders section brings balance, tier and credit details into the dashboard. Inspect the benefits recorded for the connected account.
Check receipts against a host’s work record
Receipts signed by an attested host can be checked against that host’s receipt root. Each root records the receipt key and attestation reference; it is marked anchored only after publication on chain.
Keep key rules and cached replies separate
Key policies apply to gateway requests, and cached replies are separated by account and forwarded end user. A requested cache lifetime cannot exceed the operator’s maximum.
Use the Messages API with your models
Send Messages-format requests and receive Messages-format replies or streams. The adapter uses the router’s chat path with the same key, billing, lanes and receipts.
Compare models in the Arena
Send one prompt to several models and compare answers, timing and cost. Use model presets, attested-only selection and proof badges, and choose whether to create a share link.
Inspect provider health and proof freshness
Read provider availability, attestation summaries and the history of hardware checks and probes. Proof-time observations show when checks occurred and explain their limits.
Require review before a provider can route calls
Provider applications require approval of the reviewed revision before they can serve requests. Provider credentials are protected and outbound destinations are checked.
Browse providers and their model offers
The router refreshes provider model offers, and the site shows providers with their attestation status and public registry records. An embeddable badge links a provider’s status to its evidence.
Verify receipts in your browser
Use the public verify page to inspect a receipt signature and provider attestation evidence. The page explains what each check establishes and its limits.
Read receipt and lane details from API replies
API replies carry the receipt ID, lane and applicable policy hash in response headers. Adapters pass these details through so a client can connect an answer to its evidence.
Connect tools through MCP
Connect an MCP client to list models and call the router with your key. Tools can select attested models and inspect provider evidence; calls retain the router’s billing and receipts.
Use the Responses API with your models
Send Responses-format requests, including function and freeform tool calls, through the chat router. Send the conversation on each request; stored conversations and provider-hosted tools are refused.
Keep calls on the requested lane
Choose a public, attested or unlinkable lane without a silent fallback to weaker routing. Adapter requests preserve lane errors and restrictions across their underlying calls.
Reuse model routing choices
Save model choices and fallbacks as a named route, then call it with @route/<slug>. Pin the attested lane or a disclosure ceiling; the dashboard explains when a route cannot be saved.
Read SEAL and its published specification
Browse the SEAL overview and specification on the site. Read the protocol’s evidence formats, available paths and implementation limits.
Watch spending by key and model
Inspect spending breakdowns and key budgets in Spend Watch. Set spending thresholds and webhook alerts for your account.
Add credits with Stock Tokens and track deposits
Send supported Stock Tokens to escrow and follow the confirmation wait and deposit status in the dashboard. Only final transfers receive credits, and credits are reversed if a chain reorganisation removes a transfer.
Chat with Anyroute in Telegram
Link your key to the Telegram bot and choose a model for chat. Its private mode selects the attested lane; messages still pass through Telegram and the router.
Reach Anyroute through Tor
Connect to the router’s onion service through Tor. Onion traffic uses rate limits that do not rely on the client’s network address; ordinary request text remains readable by the router.
Open account sections by link
Dashboard links keep their selected section when opened directly. Keyboard focus follows navigation so account controls are easier to reach.
Read the API reference
Browse the developer docs and the public OpenAPI reference. Find request formats, payment instructions and endpoint details in one place.
Read status without internal diagnostics
Public status responses omit private RPC credentials and background-job diagnostics. They keep the information needed to inspect service health.
Check receipt roots against chain time
Receipt roots now use chain time and strict interval boundaries. This prevents an anchor from including a receipt outside its stated window.
Manage keys and payments in the dashboard
Sign in to inspect balances, receipts and keys, and manage deposits and withdrawals. The dashboard brings these account controls together.
Browse models on the site and through the API
Browse the model catalog on the website or through the model endpoints. Read model details, provider offers and per-token prices before choosing a model.
Inspect calls and check their signed receipts
Look up call usage, cost and provider details, with signed receipts containing request and reply hashes rather than text. Published signing keys and receipt roots support verification.
Create keys with their own limits
Create API keys and set their spending and routing rules. Manage keys within the account that owns them.
Receive chat replies as they arrive
Send chat requests through an OpenAI-compatible API and receive replies as a stream. The router selects a provider and records usage for billing.
Create embeddings through the API
Send text to the embeddings endpoint using your API key. Calls use the router’s provider selection and billing.
August 2026
Pay from a prepaid balance
Deposit USDG as prepaid credits and use the balance for API calls. Credits are accounted for against your key.