Agent skill
Want the plugin instead? The one-line curl detects the agent frameworks you already have and installs Telem into as many as you select in one pass, with console retrieval metrics and trajectory analytics. This page is the lightweight path.
npx skills add https://docs.telem.aiThat installs telem-search, an agent skill: one page your
coding agent reads when a task needs the live web, teaching it to call the Telem REST API
directly. Nothing is installed into your framework and nothing reports a trajectory.
What your agent learns
Section titled “What your agent learns”The skill is a single SKILL.md. It teaches:
- When to search, and when not to. An answer already in the model’s weights, a private corpus, or a URL it already has are all reasons not to spend a search.
- The request.
POST https://router.telem.ai/v1/search, with the query inuser_inputand everything else in the optionalsearchblock — in curl, TypeScript and Python. - How to read the response. One entry per provider under
preprocessor_runs, each with its own normalized envelope,statusanderror— so one provider failing is a partial result, not a failed request. Plus the flatten-and-merge pass that turns those into one list. - How to write a query. The provider set spans two retrieval styles, so a full natural-language sentence beats a keyword fragment.
- The cost knobs.
tier,num_results, and why an explicitsearch.providers.includemakes a call’s price a decision rather than an accident. warnings[]and the error table. Which statuses to retry and which are the same failure twice.- Reading one known URL with
POST /v1/fetchinstead of searching for a page it can already name.
You can read the same page yourself: SKILL.md.
Get a key first
Section titled “Get a key first”Every request needs one. Sign in to the Telem console, create a key under API keys in your project, and put it in your environment:
export TELEM_API_KEY="tlm_..."The skill tells the agent to read the key from the environment at call time — never to hardcode it, commit it, print it into a snippet, or ship it to a browser. See Authentication.
How discovery works
Section titled “How discovery works”npx skills add https://docs.telem.ai fetches
https://docs.telem.ai/.well-known/agent-skills/index.json, finds the telem-search
entry, and downloads the SKILL.md beside it at
/.well-known/agent-skills/telem-search/SKILL.md. Hosts with their own remote-skill
registries — OpenCode among them — read the same two files.
If your host installs skills some other way, point it at either URL directly; the skill is a plain Markdown file with YAML frontmatter, with no dependencies, scripts or install step.
What you give up
Section titled “What you give up”| Agent skill | Installer | |
|---|---|---|
| Search available to the agent | yes, as HTTP the model writes | yes, as a native telem_search tool |
| Page fetch | the skill teaches POST /v1/fetch |
telem_fetch, a native tool |
| Results rendered with provider attribution | your code does it | the plugin does it |
| Retrieval metrics and trajectory analytics | no | yes |
| Touches your framework config | no | yes |
A skill costs nothing to remove and changes nothing about how your agent runs; a plugin gives the model better tools and gives you a record of what it searched for.
Related
Section titled “Related”- agents.md — the machine-readable playbook, including the migration play for code that already calls a search provider directly.
- API reference — every endpoint the skill describes.
- Configuration reference — every request option and its limit.