LLPROBeta

Home / Agents / Agent Continuity

Agent Continuity

How an agent keeps working when a service it depends on fails: check first, switch to a backup, keep the proof. The exact calls, free.

Live

What it does
How an agent keeps working when a service it depends on fails: check first, switch to a backup, keep the proof. The exact calls, free.
Who it is for
Agents and their builders.
Start in one step
Use it on this page.
What to expect
Live: open today. Free page. The tools it points to have free daily calls; their paid calls are not charged yet.

New to this? Start here.

What it does

Agent Continuity is a plan, not a tool. It tells an agent, and the person who builds it, how to keep a job going when a service it depends on fails: a model API, a cloud, a mail service or a payment rail. There are five steps, and each one comes with the exact call to an LLPRO tool that has free daily calls, on the Campus gateway (MCP at https://llpro.ai/mcp?via=agent-continuity; the same tools answer over REST). Check the service first, keep a backup provider for every kind of job, ask Backup Routing for the order to try them in, switch back when the first choice is steady, and keep signed evidence if the failure cost you money.

Five steps, each with its exact call

  1. Check the service before you change anything

    When a call to a provider fails or slows down, ask Dependency Pulse whether that provider is working right now and since when. Each reading joins the provider's own status words (provider_says) with our own timed request (we_see), by a published rule, and is signed. Read both parts: a provider can rate its whole service degraded while the part you use answers normally. Move to a backup (step 3) when your own calls keep failing and the reading agrees, for example outage, or degraded with our request failing too. If the state is up, check your own key, limits and request first. A reading is one input from one place, not a verdict on your account.

    POST https://llpro.ai/mcp?via=agent-continuity
    {"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"dependency_pulse","arguments":{"group":"models"}}}

    Returns for each service: state (up, degraded, outage, maintenance, unknown), since when, the provider's own words, our request (HTTP status, milliseconds) and a signed record. Groups: models, clouds, mail, payments, agent-rails, code; or send services (ids).

  2. Keep a backup provider for every kind of job

    Before anything fails, write down, for each kind of job your agent does, at least two providers it can use, in your order, with keys and settings ready. A backup you have never called is not a backup: run a small job on it from time to time. Look at how often each of your providers changed state. Job kinds and the providers read for each: chat-model: anthropic, cohere, google-gemini, groq, mistral, openai, openrouter, perplexity; embeddings: cohere, google-gemini, mistral, openai; email-send: mailgun, postmark, twilio; payments: base-chain, coinbase, stripe, x402-facilitator; cloud-hosting: aws-eu-west-1, aws-us-east-1, aws-us-west-2, azure, cloudflare, digitalocean, fly-io, google-cloud, supabase, vercel.

    POST https://llpro.ai/mcp?via=agent-continuity
    {"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"dependency_pulse","arguments":{"services":["openai","anthropic","mistral"],"history_hours":24}}}

    history_hours (0 to 168) lists each service's state changes over that many hours, so you can see which of your providers holds up.

  3. Ask Backup Routing for the order to try them in

    Send the job kind and your providers in your order. Backup Routing ranks them for right now by one published rule on Dependency Pulse readings (steady first, then recovering, maintenance, degraded, no current reading, outage; your order is kept inside each group) and signs the order. Use first_up, or walk the order until a call works. Do not retry the failed provider in a tight loop: each retry adds delay; move down the order instead.

    POST https://llpro.ai/mcp?via=agent-continuity
    {"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"backup_route","arguments":{"job":"chat-model","providers":["openai","anthropic","mistral"]}}}

    Returns each provider with its place, group and the reading behind it, first_up (place 1 when it is up, else none) and a signed record. Up to 10 providers; leave providers out to rank every provider read for the job kind.

  4. Switch back when your first choice is steady

    Keep working on the backup until your first choice has been up for a while, not just once. Backup Routing calls a provider steady when Dependency Pulse has seen it up for 30 minutes or more; until then it is recovering. Ask again before you move your traffic back.

    POST https://llpro.ai/mcp?via=agent-continuity
    {"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"backup_route","arguments":{"job":"chat-model","providers":["openai"]}}}

    Group steady means up for 30 minutes or more; recovering means up for less than that.

  5. Keep the proof if the failure cost you money

    If your contract with a provider offers service credits, ask Outage Evidence for a signed record of what Dependency Pulse saw for that service in the window, and send your own failed-call log lines to have their hashes countersigned as received from you. Remove keys, tokens and personal data from the lines first: they leave your agent even though only their hashes go into the record and nothing is stored. It does not file a claim, contact the provider or say a credit is owed.

    POST https://llpro.ai/mcp?via=agent-continuity
    {"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"outage_evidence","arguments":{"service":"github"}}}

    Default window: the last 24 hours. Add from and to (UTC, ISO 8601 with Z, at most 7 days back) for another window, and your_log (a list of your own log lines: at most 200 lines, 2,000 characters each, 48,000 bytes in all as JSON) to have them countersigned. Up to about 40 KB for 24 hours.

Who it is for
Agents that depend on outside services to finish a job, and the builders who run them.
Price
This page is free and has no paid tier of its own: it points to the tools it names, which carry the price. Dependency Pulse: the first 500 calls a day per caller stay free, then 0.001 USD a call. Backup Routing: 100 calls a day free, then 0.005 USD a call. Outage Evidence: 3 records a day free, then 0.25 USD a record. Those paid calls are not switched on yet: today no call is charged, and each tool's page states its current limits.
Limits
This page does not watch your agent, call any provider, hold your keys or switch anything for you; your agent makes the calls. The readings come from Dependency Pulse: one vantage point (LLPRO's server in Texas, USA), every 3 minutes, for each provider or cloud region as a whole, not your account, key, region or model. A provider can work for us and fail for you, and the reverse, so treat a reading as one input, not a promise. The steps themselves ask for nothing; the only box on this page is the site's 'Make this better' note at the foot, which keeps what you choose to send (see https://llpro.ai/privacy). The tools keep no request contents. The gateway keeps call metadata (tool, status, time, a hashed caller id, country, user agent family, referring site) for 90 days for traffic counts and abuse checks, and a running count of calls per hashed caller id with no end date yet; a task sent through the agent-to-agent door is kept for 7 days so you can fetch it.

For developers: try it

POST https://llpro.ai/mcp?via=agent-continuity
{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"dependency_pulse","arguments":{"group":"models"}}}
-> for each model API: state, since when, the provider's own words, our timed request and a signed record. Check the signature with the public key below.

For agents

Tool dependency_pulse on https://llpro.ai/mcp and in the tools list. Signed answers are checked with the public key below (signer llpro1:xigbd6eeicdufmrestnd66ewka): an Ed25519 signature over the observation as canonical JSON (sorted keys, no spaces).

Public key (base64): zCVGYxbhOFmUeBD+YDv6v7xXIG+2aykAugOQS37oKxs=

This page is part of LLPRO's agent-readable site: the same text, with the example and the key, is at https://llpro.ai/agent-continuity.md, and it is listed in https://llpro.ai/llms.txt. The tools are on the Campus gateway: MCP at https://llpro.ai/mcp?via=agent-continuity and REST at https://llpro.ai/api/v1/tools/<tool>?via=agent-continuity (POST the same arguments as JSON). The ?via tag only counts arrivals from this page; the calls work the same without it. The key on this page is Dependency Pulse's; Backup Routing and Outage Evidence publish their own keys on their pages. Errors: over MCP the answer has isError with a plain reason (unknown service, job kind or provider, a window out of range); over REST the status is 422 with the same reason.

Next

For agents: this page as Markdown · JSON · llms.txt · MCP endpoint llpro.ai/mcp · agent card