Monitors names (20 November). This is a breaking change to the Monitors API, with no
aliases for the old names. What a monitor finds is now called a result.
GET /v1/monitors/{id}/pages is now
GET /v1/monitors/{id}/results. The webhook event
monitor.hit is now monitor.result.found, and monitor.update is now
monitor.result.changed. An old event name in webhook.events returns 400. A webhook
with no events list now gets every type, results included. Before, it got no results
unless you listed them. See Webhook payload.
Results no longer have score, favicon or highlights. Monitors no longer have
agent_run_id, backfill, memory_scope, estimated_cost_usd or cost_this_month_usd,
and runs no longer have usage. The SDK method monitors.results (Go: Monitors.Results)
and the CLI command keenable monitors results need the next SDK and CLI release.Fourteen more integration pages (19 November). Coding agents:
oh-my-pi (web/keenable), OpenClaude
(/search keenable), Gajae-Code (providers.webSearch),
FrontierAgent (WEB_SEARCH_PROVIDER=keenable) and
Claude Octopus (OCTOPUS_RESEARCH_WEB_SEARCH=keenable).
Research and search: hyperresearch, career-ops,
Obscura, Qwen-MM-Plugins and the
AgentField deep research example. Also OmniRoute
(keenable-search on its search and fetch endpoints), npcpy,
Qwen Audio Agent, and one page for
RL and eval environments (SkillOpt and Halo). Halo needs an
API key; the others work without one. genoffice, OpenFilm, trueforge, Raven, Coral and weft set
Keenable up in their own settings or catalogs and are listed under
Also works with.Four more integration pages and a directories page (19 November). Setup pages for
LazyLLM (KeenableSearch), Spider
(KeenableProvider, the search_keenable feature), PipesHub
(web search provider, 1.1.0 and later) and Cerebras (the Keenable
SDK with Cerebras Inference). LazyLLM and Spider have Keenable on their main branches, ahead
of their next releases. MCP registries and directories lists
every catalog that carries Keenable, with the install command for each.Five new integration pages (19 November). Setup pages for AG2
(KeenableSearchToolkit, AG2 1.2.2 and later), OctoTools
(Keenable_Search_Tool), ToolUniverse (backend="keenable",
1.5.9 and later), tgpt (SEARCH_PROVIDER=keenable, 2.16.0 and later)
and mcporter (a keenable entry in mcporter.json). All five work
without an API key. DeepChat, DB-GPT and LocalAI set Keenable up in their own settings and
are listed under Also works with.Webhooks for MCP monitors (18 November). create_monitor takes an optional webhookUrl, a
public HTTPS address that gets each new result as a signed webhook. The response returns the
signing secret once, as webhookSecret. See Monitors tools. In the
Monitors API, an update that adds a webhook to a monitor that had
none now returns webhook.secret once. Before, it returned none, so that webhook couldn’t be verified.Four Monitors tools over MCP (18 November). With a key or a sign-in, the
MCP server lists create_monitor, list_monitors,
get_monitor_results and delete_monitor. get_monitor_results replaces whats_new and
monitor_pages: it returns the results this reader hasn’t seen yet, and with includeSeen
and a monitorId it returns one monitor’s results, read or not. A page holds up to 50
results, where monitor_pages held up to 100: follow nextCursor for more. The server no longer lists
get_monitor, update_monitor, trigger_monitor or get_run. Those actions stay in the
Monitors API, the SDKs and the CLI. A client that calls an
old tool name still gets an answer. If a scheduled task calls whats_new, change it to
get_monitor_results. See Four Monitors tools over MCP
for the old-to-new tool map and a test call.Monitors (17 November). A monitor watches the web for new pages about a topic, written
in plain words. You get each new page as a signed monitor.hit webhook minutes after it’s
crawled, in runs on a cron, interval or one-time schedule, or by polling
GET /v1/monitors/{id}/pages. Create one with POST https://api.keenable.ai/v1/monitors
and an API key. Monitors is free for now, with up to 100 active monitors per organization.
With a key or a sign-in, the MCP server adds tools such as
whats_new and create_monitor. Five new error codes cover it: monitors_not_enabled,
monitor_limit_reached, monitor_option_not_supported, idempotency_conflict and
monitors_unavailable. See Monitors.A Rust SDK. cargo add keenable installs the official Rust client. It has the same
surface as the Python, TypeScript and Go SDKs: keyless search and fetch, every search filter
and fetch option, to_context(), tool definitions with run_tool_call, and typed errors.
It is async, runs on Tokio and needs Rust 1.80 or newer. See SDKs.Exclude a whole suffix, such as .gov. exclude_domains accepts a public suffix written
with a leading dot, .gov or .co.uk, and leaves out every host under it. *.gov means the
same. A bare gov and a wildcard on a host, *.un.org, still return 400 with
invalid_filter. A suffix counts as one of the 1,000 entries. See
Exclude domains.An email before your balance runs out. When your organization has used 80% of its
monthly free allowance, so that what it has left falls to 20% of the allowance, Keenable
emails its admins once per billing period with what’s left and the reset date.
Organizations with auto top-up on get it only once auto top-up hits its monthly limit.
See Low-credit email.Plainer per-second rate-limit errors. The rate_limit_per_second error now says to
wait until the time the limit resets, then retry, and links the console page where you
can raise the limit, instead of listing top-up terms. See Rate limits.CLI: your saved key stays with the API it was saved for. The CLI sends a saved API key
only to the base URL that keenable login saved it for. With KEENABLE_API_URL pointing
somewhere else, search and fetch stop with exit code 1 instead of sending the key there.
Pass KEENABLE_API_KEY or --api-key, or run keenable login with the variable set. The
CLI also follows redirects only within the same host, reads the key from stdin with
keenable login --api-key -, and removes terminal control characters from page text in
-p output. See Saved keys and other base URLs.Safe search for pro and realtime search. Search takes safety_mode: off (the default),
moderate or strict. With moderate or strict, results that a safety model judges
unsafe at that level are removed, and strict removes more. Removed results are not
replaced, so a query can return fewer results, or none. safety_mode works only with
mode set to pro or realtime. In another mode, with disable_cards, for a URL query, on
Time Machine search, or with any other value, the request returns 400 with
invalid_parameter. See Safe search for a sample request
and each error, and the Search request.See your credits left on every call. The usage object on /v1/search and
/v1/fetch, and _meta["keenable/usage"] on MCP tool calls, has a new field,
credits_remaining: your organization’s free allowance and purchased credits left after
the call. Watch it to see the allowance running out before calls fail with 402. See
Credits.MCP: shorter fetches by default, and the rate limit in the instructions. The MCP
server’s fetch_page_content returns at most 20,000 characters when you don’t set
max_chars, down from 50,000, so a long page no longer floods an agent’s context. The
Truncated: line gives the full length; raise max_chars or use prompt for more.
The server’s instructions now state the rate limit, tell agents to send at most 5 calls
at once and to wait out a rate-limit error’s retry time, and describe the less obvious
options (max_chars, live, snippet_max_length, max_results, output_format,
session_id). See MCP server.CLI 0.5.1: an exit code for each kind of failure. keenable search and keenable fetch
exit with 3 for authentication, 4 when rate limited, 5 when out of credits, 6
when the page or site failed, 7 on a network failure, 8 for an invalid request and
9 for a server error worth retrying, instead of 1 for all of them. In YAML mode, a failure also prints a one-line summary
to stderr. Scripts that only check for a nonzero code keep working. With -p, the CLI
shows your credits left. See CLI exit codes.Leave sites out of search. Search takes exclude_domains, a list of up to 1,000
host names such as ["pinterest.com", "quora.com"]. Results from those hosts and their
subdomains never come back, so fewer than max_results can return. An entry that isn’t
a host, such as a path or a wildcard, returns 400 with invalid_filter and names the
entry. The MCP server’s search_web_pages tool takes the same parameter. See
Exclude domains.Search and fetch refuse parameters they would misread. A parameter you send now
either takes effect or the request fails with 400. Fetch returns unknown_parameter
for a parameter it doesn’t know, as search does, instead of ignoring it. live takes
only true or false, and max_results and max_chars take only integers in their documented
range, sent as a number or a string of digits, so live=1 and "max_results": true fail instead
of being read as something else. When
an unknown name is close to a supported one, such as count for max_results or q
for query, the message asks Did you mean and the new did_you_mean field maps it
to the parameter to use. A POST body that isn’t JSON now says so instead of
reporting "query" is required. If your client sends extra parameters to fetch, or
live=1, update it. See 400 Invalid request parameters.Untitled results get a title from the URL. A search result whose page has no title
used to come back as Untitled. Its title is now built from the URL’s host and path,
such as api.github.com/repos/vercel/next.js/releases/latest, so your model can tell
such results apart. The MCP fetch tools label an untitled page the same way. See
Search.Fetch dates match search. Fetch now returns published_at as an ISO 8601 timestamp
in UTC, such as 2026-10-30T00:00:00Z, the same format as search results, instead of
Unix seconds. This covers /v1/fetch, the public, x402 and ZeroClick endpoints, and the
MCP server’s fetch tools. The SDKs ignored the number, so they now return the date.
Code that parsed the number should accept the string. See Fetch.CLI 0.3.0: session IDs, fetch dates and clearer errors. keenable search and
keenable fetch take --session-id to group the calls of one agent run. Fetch output
shows published_at and acquired_at as RFC 3339 timestamps in UTC, like search, and
adds truncated: true when the content hits the --max-chars cap. Errors carry the
API’s request ID and a hint with the next step, plus type, code, fix and
docs_url when the API sends them. KEENABLE_API_URL sends API calls to another base
URL, such as a local gateway. See Session IDs,
Truncated content, Errors and
API base URL.CLI 0.3.0 on corporate networks. The CLI trusts your operating system’s certificate
store, so a company root CA installed on the machine works without setup. It also reads
SSL_CERT_FILE and SSL_CERT_DIR, sends requests through HTTPS_PROXY and reaches
hosts in NO_PROXY directly. Add -v (--verbose) to see the proxy and certificates
in use. See Proxies and corporate networks. To
upgrade, run keenable update, or brew upgrade keenable-cli for a Homebrew install.MCP server: clearer errors and server info. Tool errors now say how to fix the
call: a bad mode names pro and realtime, an unknown tool lists the available tools,
an internal error carries the request id, and an invalid API key on /mcp says to check
it or drop it to search without a key. The date filters’ descriptions now mention that
they take ISO timestamps and relative offsets such as 7d, and initialize returns a
title, description, website and icon for client UIs. See MCP server.Search without cards. For some queries, such as prices, exchange rates or facts
about a company, search returns a card: a result that Keenable writes itself, at
https://cards.keenable.ai/w/<id>. Send disable_cards: true to get ordinary web pages
in their place. disable_cards works in pro and realtime modes, on the
Search API and the MCP server.Pay per request with x402. Search and fetch can now be bought one call at a time,
with no account and no API key.
/v1/x402/search and /v1/x402/fetch answer 402 with
an x402 offer. The client pays $0.01 in USDC on Base, retries,
and gets the result with an on-chain receipt. Any standard x402 client works. See
Pay per request.Auto top-up. Keenable can now refill purchased credits automatically. Set a threshold, an
amount and a monthly limit under Billing in the console, and Keenable charges your saved
card whenever your balance falls below the threshold. If the card is declined, Keenable
turns auto top-up off and emails you, and doesn’t retry the charge. See Auto top-up.Built into Hermes Agent. Keenable ships in Hermes Agent
as a bundled web backend for web_search and web_extract, with no plugin and no key.
Run hermes config set web.backend keenable to select it. You no longer need the
standalone hermes-keenable-web plugin.Versioning and deprecation policy. A new policy sets out what can change within /v1,
what counts as a breaking change, and at least 90 days’ notice by changelog and email before anything you
rely on stops working. See Versioning and deprecation.In the Plugin marketplace in Antigravity. Keenable is one of the launch plugins in
Google Antigravity’s new marketplace: install it from Customizations → Marketplace
for web search and page fetch, with no key. See
Google Antigravity.Keenable on Baseten. Web search and page fetch run as
server-side tools on Baseten Model APIs. Name the two tools in a
chat request and the model searches and reads pages inside Baseten’s inference loop. You
don’t need an MCP client or a Keenable key, because Baseten meters the calls onto your
Baseten invoice.Keenable SELECT in both directories. The SELECT server is now a listed
connector in Claude (Settings → Connectors → Keenable SELECT) and a published plugin in
OpenAI’s Plugins Directory, shared by ChatGPT and Codex. Both sign you in with your
Keenable account, so there’s nothing to paste. The custom-connector and developer-mode routes keep
working for accounts the listings have not reached yet.
Point-in-time search. The
query_time parameter is now documented on both the
MCP tool and the Search API: search the
index as it stood at an instant, and have relative date filters resolve against that
instant rather than now.Two more search parameters. max_results caps how many results come back (1–50),
and snippet_max_length caps the snippet per result. Both are documented on the
MCP server and Search API pages.Integrations directory. A catalog of the frameworks, agents and
editors Keenable plugs into, with one-click installs where the client supports them.
Added this month: Claude, Convex, OpenCode, Pipecat, the Vercel AI SDK tool package,
Lightpanda, LibreChat, and a listing in the OpenAI Plugins Directory.Keyless endpoints, written down. /v1/search/public and /v1/fetch/public are
documented in the API reference. Use them to evaluate the API
without an account. Use an API key in production.Credits, on their own page. The monthly credit allowance and the per-SKU pricing
model moved out of Rate limits into Credits.Prompt-driven extraction. The Fetch endpoint accepts a
prompt, and returns what an LLM read out of the page for that instruction instead of
the whole page.Usage and overrides over MCP. Every authenticated tool call now reports its cost
under _meta["keenable/usage"] (reading usage per call),
and an integrator can pin the search mode with _meta["keenable/overrides"]
(operator overrides).Date-only cutoffs cover the whole day. A bare YYYY-MM-DD on acquired_before /
published_before now resolves to 23:59:59.999 on that date, so pages from the named
day are kept at either end of a window. Pass a timestamp to cut at an exact instant.Live fetching on the Fetch API. The Fetch endpoint now
accepts a
live parameter to retrieve a page in real time instead of serving the
cached copy. Use it when freshness matters more than latency.Published rate limits. Request limits are now documented: 10 requests/second per
organization for authenticated API keys, and a 1,000 requests/hour quota for
unauthenticated use. See Rate limits.One-click MCP install. Added one-click install buttons for Cursor and VS Code
(including VS Code Insiders) so you can add Keenable to your editor without editing
config by hand. See MCP server.Unauthenticated access. The CLI and MCP server now answer
requests without an API key, on a shared public tier, so you can evaluate them before
creating an account. See Authentication.Authentication and rate-limit guides. New Authentication and
Rate limits pages cover the device-code login flow, API keys, and the
unauthenticated vs. authenticated tiers.
Keenable API launch. The initial public release of the Keenable platform for AI
agents: the Search API (
/v1/search), the
Fetch API (/v1/fetch), the CLI, and the
MCP server, with one API key across all of them.