Command catalog
The complete Metrecept surface in any connected host: ten MCP tools, each
mirrored by a slash skill in Cursor where applicable (/ohm-chat,
/ohm-fetch-web, /ohm-receipt, /ohm-inventory, and so on). Connect a host
first — see Integrations or /connections.
| Tool | When to reach for it |
|---|---|
ohm_chat | Route a model call through the pipe so identical prompts replay from cache |
ohm_fetch_web | Pull public web pages into context, compliantly |
ohm_usage | Check your meters: cache hits, fetches, pipe rent |
ohm_savings | See what the cache has saved you |
ohm_receipt | Mint a public savings receipt (/r/…) |
ohm_models | List the model ids the pipe routes to |
ohm_providers | Check upstream provider and failover health |
ohm_policy | See which web-fetch purposes are allowed |
ohm_cache_trees | List your cache-tree inventory |
ohm_cache_inventory | List cached entries in a tree — digest, model, TTL, size |
ohm_chat
Chat through the pipe (OpenAI-compatible). Identical prompts replay from the Redis cache instead of hitting the provider again.
Example prompt to your agent:
| Param | Default | Notes |
|---|---|---|
prompt | (required) | User message |
model | mock | mock works without keys; gpt/claude need BYOK |
fetch_urls | none | Optional public URLs attached as compliant web context |
purpose | public_web_retrieval | Applies when fetch_urls is set |
upstream_api_key | "" | Per-call BYOK override (else OHM_UPSTREAM_KEY) |
Use ohm_chat to summarize https://example.com/changelog with model mock.
ohm_fetch_web
Compliant public-web fetch: robots-gated, PII-redacted, metered as ohm_web_fetch. Public pages only.
Example:
| Param | Default | Notes |
|---|---|---|
urls | (required) | Public http(s) pages |
purpose | public_web_retrieval | Also: business_catalog, public_company_info, job_listings |
query | "" | Optional focus question for the summarizer |
format | markdown | markdown or json (title, text, meta, JSON-LD) |
Use ohm_fetch_web on these three docs pages, format json, query "auth setup".
ohm_usage
No params. Returns cache hit ratio, fetch counts, and estimated pipe rent from GET /v1/usage.
Call ohm_usage and tell me my cache hit ratio this cycle.
ohm_savings
No params. Returns replayed prompt counts and estimated spend avoided from GET /v1/savings. Pair with ohm_usage for the full meter picture.
Call ohm_savings — how much has the cache saved me?
ohm_models
No params. Lists the model ids the pipe routes to from GET /v1/models, including BYOK upstreams. Use it before picking a model for ohm_chat.
Call ohm_models and pick the cheapest model that can handle this task.
ohm_providers
No params. Returns upstream provider and failover status from GET /v1/providers. Check this first when a real (non-mock) model call errors.
ohm_chat is failing on gpt — call ohm_providers and check upstream health.
ohm_policy
No params. Returns the allowed web-fetch purposes and their limits from GET /v1/compliance/policy. Use it to pick the right purpose, or to explain a refused fetch.
Call ohm_policy — is job_listings an allowed purpose on my seat?
ohm_cache_trees
No params. Lists your cache-tree inventory from GET /v1/cache/trees
(tree_id, parent_tree_id, status, default, created_at).
Call ohm_cache_trees and tell me which tips I have besides main.
ohm_cache_inventory
Lists individual cached entries in one tree from
GET /v1/cache/trees/{tree_id}/entries — digest, model,
ttl_seconds_remaining, total_tokens, size_bytes. Metadata only; never
returns cached message content.
| Param | Default | Notes |
|---|---|---|
tree_id | main | Cache tree id |
limit | 25 | Capped at 100 server-side |
offset | 0 | Pagination offset |
Call ohm_cache_inventory on tree pr-142 and tell me what's still live.
ohm_receipt (and demo mint)
Mint a public savings receipt (POST /v1/savings/receipt). The /demo
Mint public receipt button after miss→HIT is the same mint — no MCP required.
Receipts are estimate_only (display name + aggregates; never prompts).
Path + hit ratio
Send X-Ohm-Path (or JSON ohm_path) on chat to label a frequency farm. Read
inventory with GET /v1/ledger/hit-ratio (tenant) or
GET /v1/org/ledger/hit-ratio?month=YYYY-MM&group_by=path|cost_center (org).
Environment
Next: optimized usage — the workflow that gets the most out of these ten.
| Var | Role |
|---|---|
OHM_API_KEY | Tenant key (sk-at-…) — required |
OHM_BASE_URL | Defaults to https://api.withohm.dev/v1 |
OHM_UPSTREAM_KEY | Optional BYOK provider key for cache misses |