FluxA AgentMarket vs Tempo

FluxA AgentMarket 12

Graded as a hosted service, the discovery gateway on agentmarket.fluxapay.xyz and the proxy and LLM endpoints on router.fluxapay.xyz, from the vendor's pages. Nothing was called. No status page is linked from the site, the footer, llms.txt or the guides, and no status host was checked (0). With no readable incident history the record takes the default (5). Listing pages show a health count with no history, such as '10/10 endpoints healthy' on Minifetch. No rate limit with numbers was found in llms.txt, the guides or the CLI help. The only cap is a per-key spending limit (0). The Balance guide says to retry a 503 shortly without minting a new credential, replays of a settled top-up finalize call are idempotent, a failed order answers 409, and the card flow says 'a retry is a new order'. No 429 or Retry-After guidance was found. Termix earned 8 for the same kind of guidance with idempotency keys on writes (7 of 15). No SLA was found, and the terms supply the service 'as is' (0). The about page labels the product 'v.01 / Beta' (0). Total 12.

Tempo 66

Status page at status.tempo.xyz with nine components across mainnet, testnet and the Tempo API (20). The page moved from Better Stack to incident.io, and on 1 November its history starts on 12 October 2026, with every component at 100% and no incidents listed since then. For 3 August to 12 October we rely on the old Better Stack feed as read on 17 October, which showed only short public RPC outages in the window, mainnet on 28 September (09:15 to 09:19 UTC) and 6 October (00:47 to 00:51 UTC), and testnet on 5 August, 23 September and 12 October (00:08 to 00:15 UTC), each under 15 minutes, so minor only (20). The rate-limits page, unchanged since 9 October, says "Limits depend on the endpoint, key configuration, and billing state" and asks callers to read RateLimit-Limit, and the OpenAPI still gives 100 requests a minute for anonymous MCP calls. Quota structure, RateLimit-* headers and one published number, short of the checklist's limits documented with numbers (10 of 15). 429s carry Retry-After, and the docs ask for backoff with jitter and give a five-attempt helper; we found no idempotency guidance for the fee-payer relay (13 of 15). No SLA (0). Mainnet has run since 18 March 2026, but the versioning policy still says "Endpoints are not yet stable and may change without notice" on 1 November (3 of 10). 20+20+10+13+0+3 = 66.

FluxA AgentMarket 53

No OpenAPI file was found. llms.txt names none and the documentation site, docs.fluxapay.xyz, couldn't be read because its robots.txt couldn't be. The discovery response has documented fields (apiServers, skills, models, pagination), a GET on each listing's urls.access returns its endpoints and prices as JSON per the skill-creation guide, and the LLM endpoints follow the OpenAI and Anthropic formats (8 of 25). llms.txt and Markdown guides (getting-started.md, currency.md, create-skill.md, the APIs and Balance guides) written for agents (10). The guides say when to use search against query, that the external catalogues are a last resort and which rail fits which job. Listing descriptions come from creators and 17 of the first 50 skills have none (12 of 20). type is an enum that answers 400 to unknown values, limit is bounded at 1,000, and the chat and messages bodies follow published formats. Listing endpoints are described by each creator (9 of 15). curl examples for discovery, models and payment, a status table in the Balance guide with the action for 401, 503, 404, 402 and 409, and seven named mandate errors in the APIs guide (11 of 15). Paths carry /llm/v1/ and llms.txt names minimum CLI versions, but no changelog was found (3 of 15). Total 53.

Tempo 90

Public OpenAPI 3.1.0 at api.tempo.xyz/openapi.json, read on 1 November, with 136 paths and 159 operations (129 paths on 17 October), plus OpenRPC for JSON-RPC (25). llms.txt with 339 links and Markdown twins (10). Per-domain API guides, and the two MCP servers are told apart, mcp.tempo.xyz for documentation and api.tempo.xyz/mcp for the API, whose meta-tools say when to inspect a tool before calling it (16). Typed parameters such as chainId (mainnet, testnet or a number) and limit bounded to 5 to 200 (12). A single error envelope with stable error.code, field paths on validation errors and a request ID, plus a full code catalogue (14). /v1 paths, a versioning policy with Deprecation and Sunset headers that applies once the API stabilises, and a dated node changelog (13).

FluxA AgentMarket 32

The editorial half. The marketplace, proxy and gateway are closed. The terms (effective 14 January 2026) cover pricing, refunds and acceptable use but name no governing law, termination terms, liability for third-party listings or creator obligations, and the company appears only in the footer. The CLI is Apache-2.0 (8 of 30). The privacy policy (14 January 2026) lists account, payment and usage data and says identity isn't shared with model providers. It says nothing on prompts, API payloads, retention, training or data location (6 of 30). No deprecation policy or dated notice was found, and the terms may change 'at any time' (0 of 20). The policy names Stripe and 'OpenAI/Anthropic API' as a non-exhaustive list, while the models page lists AIHubMix, UnifyLLM and Baidu AI Cloud as providers. No locations are given (4 of 20). Total 18. The provenance half is computed from the provenance block.

Tempo 68

The node is MIT or Apache-2.0, but the API, console, CLI and MCP server have no terms of service we could find; tempo.xyz/terms and tempo.xyz/terms-of-service both returned 404 on 1 November (20). The privacy policy, still last updated 31 July 2026, names Tempo Labs Inc. dba Tempo, keeps data "as long as necessary" with no periods, notes that on-chain data can't be deleted, and covers transfers to the United States under Standard Contractual Clauses; the docs say API tokens are hashed at rest and redacted from logs. No DPA found (15). A written versioning and sunset policy, and every network upgrade in the changelog carries testnet and mainnet activation dates. T12 shipped in v1.16.0 on 6 October for testnet on 8 October and mainnet on 13 October. T13's page went up on 16 October, its dates were added on 22 October, and it activated on testnet on 29 October with mainnet set for 3 November, so 7 and 12 days' notice. Notice has been as short as three days (16). Stripe is named for console billing and PostHog for web analytics, and the policy says data may be processed in the United States; no subprocessor list found (6).

FluxA AgentMarket 66

Graded on the HTTP API. Discovery takes limit up to 1,000, type and tags filters and a Markdown format, with no field selection, and listing responses vary by creator (15 of 25). limit, offset and nextOffset on discovery, search and query modes, and the ledger takes kind, limit (maximum 100) and offset. llms.txt says filters ignore limit and offset (16 of 20). The Balance guide's status table gives an action for each status, the APIs guide names insufficient_balance, mandate_budget_exceeded, agent_not_registered and others, and a model served only on Messages answers 400 on chat (15 of 20). finalize replays are idempotent and a dead order answers 409 with the instruction to start a new one. No idempotency key on paid calls was found (10 of 20). The LLM endpoints accept any OpenAI or Anthropic client, discovery needs no parameters, and the only official client is the TypeScript wallet CLI (10 of 15). Total 66.

Tempo 94

api.tempo.xyz/mcp lists four compact meta-tools (search_tools, get_tool_details, call_read_tool, call_write_tool, about 2.3 KB in all) that page through 92 API and docs tools on 1 November (87 on 17 October) and load a schema only on request, and mcp.tempo.xyz lists four documentation tools (25). Cursor pagination on every list endpoint, optional page pagination, and limit from 5 to 200 (18). Errors tell the agent to branch on error.code and point at the offending field, and the 402 for payment is protocol-native (18). Every MCP tool on both servers carries readOnlyHint and destructiveHint, writes go through a separate call_write_tool, and the OpenAPI takes an Idempotency-Key on five write operations on 1 November (route transfers, vault and zone transfers, V2 deposit addresses and MPP broadcast; the separate universal deposit address operation listed on 17 October is no longer in the file), with a 24-hour replay window on route writes, though not on webhook writes (18). Public reads need no key and default to mainnet, with clients for TypeScript (viem/tempo, mppx), Python (pympp) and Rust (15).

FluxA AgentMarket 64

Read for a closed service. No changelog or dated release notes were found. The models page says the catalogue was 'last updated November 5, 2026', and @fluxa-pay/fluxa-wallet 0.6.0, the CLI that carries the market commands llms.txt relies on, was published on 6 November 2026 (30). The CLI had eight releases between 10 August and 6 November 2026, 0.4.12 to 0.6.0 (20). Support is a Discord server and support.fluxapay@fluxapay.xyz, neither tested, and GitHub issues couldn't be read because the GitHub API refused us (3 of 15). No AgentMarket SDK. The TypeScript CLI is the official client and the LLM endpoints take OpenAI and Anthropic SDKs. Not an MCP server, so the registry line reads as SDKs (7 of 15). The CLI repository has 7 test files and no CI workflow, and the npm package declares no repository. @fluxa-pay/fluxa-connect-mcp points to a repository that isn't public (4 of 10). Total 64.

Tempo 88

Node v1.17.0 (T13) published in the changelog on 26 October 2026, tagged tempo/v1.17.0 on 23 October (30). Ten releases from v1.12.0 on 3 August to v1.17.0 (20). 175 commits to the node repo's default branch since 17 October and a dated public changelog; we didn't read the issue queue (18 of 25). Current TypeScript, Python and Rust clients, with pytempo and tempo-go 0.6.1 listed for T12; we didn't check the MCP registry (10). CI workflows for tests, lint, semver checks, dependency scans and reproducible builds, re-read in the repository on 1 November (10).

FluxA AgentMarket 51

fxa_live_ keys are created by a person at /keys and shown once, each with a spending limit in Units that changes within seconds, and are revoked immediately over DELETE /llm/keys/<id>, with written rotation steps. Key management needs a 5-minute Agent VC, never a key, and creation can be set to need a person's approval. No permission scopes beyond the spending limit were found, and no secret in a URL (24 of 30). Wallet payments need a mandate a person signs with a USDC budget and an expiry, and the Balance guide says to confirm every top-up with the user. The APIs guide tells agents not to ask before paying within a signed mandate, and no read-only key was found (14 of 20). The marketplace returns content from third-party APIs and installs third-party skills. Results carry a fluxa or bazaar source tag, but entries named ssrf1, keyprobe and tooltest1 are tagged fluxa, no review of uploads is described, and no guidance on treating responses or skills as untrusted was found (2 of 15). The balance endpoint returns available, spent, burn per day and owed, the ledger lists spend and purchases per merchant, and the wallet skill reads an X-LLM-Cost header on model calls (11 of 15). No disclosure policy, bug bounty, certification or advisory was found. security.txt couldn't be read and is scored absent, and the privacy policy says only 'industry-standard security measures' (0 of 20). Total 51.

Tempo 60

Project-scoped API keys with named scopes such as data:read, documented rotation and revocation, optional IP allowlists, environment prefixes (tempo:sk: and tempo_sandbox:sk:), and only a hash stored at rest. MPP payment credentials are kept separate from keys (30). The OpenAPI still lists a key query parameter as a "Supported query-string alternative for API keys" on 1 November, with a warning that URLs may be logged, so a secret can travel in a URL (minus 10). Sandbox keys can't touch mainnet, most endpoints are read-only, the API MCP server splits read and write calls, and the console sets monthly spend and fee-sponsorship limits. Payments are signed by the agent's own wallet with no approval step on Tempo's side (16). Chain data such as token names and memos is attacker-controlled and we found no prompt-injection guidance (6). Console usage reports, Payment-Receipt headers and a public ledger (10). The organisation-wide SECURITY.md (in tempoxyz/.github, shown on the node repo) asks for reports to security@tempo.xyz with an acknowledgement within 5 days, and it and the node README still say Tempo "is still undergoing audit and does not have an active bug bounty". tempo.xyz/.well-known/security.txt returned 404 on 1 November. A security release (v1.13.1, 20 August 2026) was announced in the public changelog, the node patched rustls advisories in its dependencies on 14 October and RUSTSEC-2026-0385 (ctutils) before v1.17.0, and bumped tokio for RUSTSEC-2026-0386 on 23 October, a dependency fix in no tagged release on 1 November (8). Tempo doesn't hold user funds; wallets are self-custodied and stablecoins are issued by third parties.