Amdahl Platform API

The user's resume payload. Must match the run's persisted pending_input_schema.

Optional operator note describing why the run was canceled.

Optional stable kebab-case dispatch key (lowercase letters, digits, single dashes). Omitted: derived from the name and uniquified automatically.

Display name.

Optional one-line blurb for the agent directory.

Operation ids to subtract from this agent tool kit at run time (all kit tools are on by default; empty list = no restriction). Runner-inline tools are refused.

The prompt the agent runs with.

New display name.

New directory blurb.

New prompt body.

Operation ids to subtract from this agent tool kit at run time (all kit tools are on by default; empty list = no restriction). Runner-inline tools are refused.

Human label. Each fired Chat is named "{name} — {date}".

The Chat input every fire starts the Master turn with.

Cron expression (e.g. "0 9 * * 1"). Validated before the row is stored.

IANA timezone the cron evaluates in. Defaults to UTC.

LEGACY single-agent pin the Master turn runs as (library slug/uuid or workspace agent id/slug). Prefer config.agents; passing both rejects.

Per-fire Chat config. on_question is always none (headless); write_outputs defaults true for routines; agents picks the roster ("all", one pin, or a delegate allowlist).

Whether the routine fires. Defaults to true.

New human label.

New Chat input for future fires.

New cron expression (validated before storing).

New IANA timezone.

LEGACY single-agent pin, or null to clear back to the plain Master. Prefer config.agents; passing both rejects.

Replacement per-fire config (replaces the whole object, including the agents roster).

Enable or disable firing.

The ask, in plain language. Becomes the Master turn user message.

Continue an existing Chat. Omit to open a new one.

Name for a NEW Chat. Omitted: auto-titled from the input.

Optional agent pin: an Amdahl library slug/uuid or workspace agent id/slug the Master turn runs as (Master privileges; the agent prompt is the specialization).

The only knobs. No model, temperature, or stream fields exist.

The new chat name.

Default Chat run knobs. Replaces the stored run_config wholesale when sent.

Agent slugs / refs the Master may NOT delegate to for this user. Empty = every agent usable.

Free-text ask. Meaning-shaped wording ("what do customers say about…") routes to the semantic lane; other questions route to the NL-to-SQL fuzzy lane.

How many independent sub-questions a multi-part ask may fan out into. Default 3; up to 5 synchronously and 12 with async:true. RAISE THIS when a reply comes back with retry_guidance.action = "raise_max_subqueries" — it means parts of your question were cut by this cap, and retry_guidance.suggested_params carries the value to use.

Run as a background job instead of blocking. Returns {job:{job_id,status}} straight away; collect it by calling this verb again with that job_id and no query. USE THIS when a reply comes back with retry_guidance.action = "run_async" — the synchronous path is bounded at ~15s and your ask needs longer, and the async path also allows a higher max_subqueries. Prefer it for any deliberately broad, multi-part question.

Collect a job started with async:true. Send this ALONE (no query). While status is "queued"/"running", poll again; on "complete" the full envelope is on detail.

Lane override. Default auto: filters-with-no-query → filter; meaning-shaped query → semantic; else fuzzy. Lanes read DIFFERENT stores at different grains — read `corpus` on the response to see which one answered.

Warehouse surface for the filter lane. Default interactions. Ignored by the fuzzy and semantic lanes, which report it back in corpus.ignored_params when you pass it.

Who counts as the voice you want. Default "all". Use "customer_voice" to keep only genuine customer / prospect speech and drop investor, advisor, media, partner, vendor and competitor chatter that merely shares an account record — while STILL keeping a real buyer on such an account (someone on a live opportunity, or a scored champion / economic buyer). This applies the platform's own canonical rule server-side; it is a disjunction you cannot express with `filters`, which are ANDed. interactions only.

Semantic lane only. Semantic matches carry a truncated preview and no labels, so they cannot be quoted or attributed. Set true to bridge each match back to its full warehouse row — full `content` plus company, speaker title, pushback and buyer-signal columns. One row per match, containing the matched utterance, so a hit is citable. SPEAKER comes from the ATOMIC utterance that matched, not from the turn stamp: read speaker_attribution on each row. "utterance" = verified from the atomic source. "ambiguous" = two matched utterances share this row and disagree on speaker, so the speaker fields are NULL and nobody is named — check turn_speakers for who is in play. "turn" = the atomic source had no row, so the fields are the turn's FIRST member and are NOT verified. Separately, speaker_ambiguous says the row's TEXT spans more than one speaker (13.58% of turns on the tenant measured 2026-08-15) — a row can be cleanly attributed AND span two voices, so quoting the whole content as one person is still wrong there. Both fields ship on every turn-grain row including the clean ones; corpus.speaker_ambiguous_rows / corpus.speaker_unattributable_rows are the rates. speaker_ambiguous_source says where that verdict came from: "stamped" = read from the column the pipeline writes on the turn itself, "overlay" = reconstructed at read time because the pipeline recorded none for that turn. A missing record is NOT "one speaker" — permanently-excluded interactions never get one — so an unrecorded turn falls back rather than being reported unambiguous, and a "stamped" row also carries speaker_count. corpus.speaker_stamped_rows / corpus.speaker_fallback_rows are the split. Stakeholder columns (role_level, is_champion, is_economic_buyer, champion_score, inferred_persona) are derived from the turn stamp, so they are NULLED when the row is re-attributed to a different speaker rather than left describing someone else. Slower — two reads. Check corpus.hydrated_grain: "turn" is that row-per-match answer; "call" means the turn lookup failed and the rows are arbitrary turns from the right conversations (context, not the matches; also flagged corpus.degraded). Matches whose utterance belongs to no turn have no warehouse row and are counted in corpus.unresolved_matches; matches sharing one turn are corpus.collapsed_matches. Read those two directly rather than inferring a gap from a match_count/row_count delta — corpus.match_count is how many nearest neighbours the lane pulled and is BOUNDED BY limit, not a count of how many things genuinely matched. Caps at 200 distinct ids, reported as corpus.truncated.

Opt into gap detection on corpus.coverage. Every response ALREADY carries corpus.coverage with max_timestamp / min_timestamp and an age histogram, computed free from the rows returned. This flag adds `gaps`: windows where the corpus HELD rows and this result set had none — the "your evidence has a hole in it" signal, as opposed to the ordinary quiet periods a plain empty-bucket scan would flag. Costs one extra bucketed read, so it is off by default. Read coverage.basis first: "matched_set" means the numbers describe your whole match set, "returned_rows" means they describe a truncated slice, and "unavailable" means coverage could not be measured — never read that as zero.

Typed predicates (ANDed — there is no OR; for the customer-voice disjunction use `audience`). Semantic mode supports only company_id / occurred_at / speaker_type, and cannot apply any of them on its theme-index fallback — check corpus.filters_applied. The filter lane supports every field the catalog advertises.

Filter lane: order by a vocabulary field or a metric alias (e.g. "count").

Row / match cap; hard max 1000. Default is lane-dependent: 100 on the filter and semantic lanes, 50 on the fuzzy lane (where an unforced plain-language question usually lands), so pass it explicitly when the row count matters.

Filter lane: group fields (requires at least one metric).

Filter lane: aggregations (sum/avg need a numeric field).

Eval slug or id. Default 'prompt-and-message-eval'. Discover options + their input fields via eval://list.

Run inputs for the eval's declared fields. prompt-and-message-eval requires at least one of prompt / message, and also accepts audience (a seniority or title to scope the cohort), account (the company it is going to — retrieves what THAT account said), and mode: rewrite (the default, a full improved prompt and message), advisory (anchored suggestions against what you wrote), or gate (grade ONLY what you sent and stop — read the verdict from eval-run://<id>/gate). Read the exact schema from eval://<slug>.

cached (default) reuses an active run for the same target (fingerprint); force always starts a fresh run.

Run id whose frozen customer quotes this run should be graded against, instead of retrieving fresh ones. Use to A/B two drafts: without it each run retrieves its own evidence, so the score difference mixes your edit with a different set of quotes. The source run must be in this workspace and have reached retrieval; the report states the pin and how old that evidence is.

Add public web/market evidence to the grade (default false). One web fan-out runs and its snippets join the pool tagged EXTERNAL — the tier that backs market claims ("analysts report"), never customer voice. Needs the external_search:execute scope (refused without it, never silently dropped) and cannot combine with a pinned run — a pin holds the evidence frozen, and its pool already carries any external quotes it was graded on.

How many external quotes to retrieve when include_external is set. Clamped to 1..25; default 15. The ack echoes the clamped value as external.cap.

Run id whose improved version this run should GRADE, instead of writing a new one. Use to re-run the judge over fixed text — the run makes no generate call, so a score difference between two such runs is the judge, not the writer. Pass reuse=force when sampling: two identical pinned requests share one content address, so on the default reuse=cached the second is served the first run and an N-sample spread reads as zero. Also pins that run's quotes (the citations in the frozen text address them), so pass the same id to evidence_from_run or omit it. Refused if this run submits different copy or a different mode than the pinned candidate was written for.

Narrow WHICH conversations the grade speaks for. Use when the verdict is about a cohort rather than your whole corpus. filters are AND-joined over interactions (utterance grain), deals or deal_qualification (account grain - these resolve to the matching accounts and scope the conversations to them). A slice below the evidence floors (3 speakers, 25 utterances, 2 companies) abstains to whole-corpus grading unless allow_thin_evidence, which grades it anyway and flags below_floors.

kebab-case identifier, unique in your workspace.

semver, default 1.0.0.

Declared run-input fields ({ fields: [...] }).

Cases: each a subject + its graders.

Required for any grader that reports a MAGNITUDE (judge / evidence_judge / improvement_loop): the measured flip rate and the floor a reported lift must clear. Validation refuses without it, so declare it rather than discovering the refusal.

Required for any grader that reports a MAGNITUDE (judge / evidence_judge / improvement_loop): the measured flip rate and the floor a reported lift must clear. Validation refuses without it, so declare it rather than discovering the refusal.

Required for any grader that reports a MAGNITUDE (judge / evidence_judge / improvement_loop): the measured flip rate and the floor a reported lift must clear. Validation refuses without it, so declare it rather than discovering the refusal.

Eval slug or id. Default 'prompt-and-message-eval'. Required in effect — columns and versions are per-eval, so one export covers one eval.

Restrict to these eval_version values. Omit for every version, but read the version spread on the count first: scores from different versions are produced by different instruments and are not comparable.

ISO-8601 lower bound on created_at (inclusive).

ISO-8601 upper bound on created_at (exclusive).

Restrict to these run statuses. Omit for every status.

Restrict to these verdict buckets. Omit for every bucket. `ungraded` is a pseudo-bucket for runs carrying no verdict at all - the ones whose overall_score cell is empty - so you can isolate or exclude them; it is not a value any run stores.

Return only the count block (total, cap, version spread, date range) without the rows. Cheap enough to call on every filter change.

Blank the verbatim customer quotes in the exported verdict. Affects the raw JSONL download only; the flat rows never carry quote text.

Proceed when more than 5000 runs match, taking the most recent 5000. Without it an over-cap response is refused rather than silently cut. The .csv / .jsonl file downloads stream and have no cap, so they never need this.

Take at most this many runs, most recent first (max 5000 here). A deliberate top-N, not a truncation: it satisfies the cap on its own, so a limited call is never refused for matching a larger set.

The run this feedback is about.

What happened to the recommendation. `unsure` is a real answer and is excluded from both halves of any adoption rate — prefer it to guessing.

How you know. `did_it_myself` = you performed the action in this session (the only value that can be verified). `reported_to_me` = a person told you. `inferred` = you concluded it; honest, but excluded from published rates. Never claim first-hand knowledge of something a person did outside your session.

What the feedback is about. Omit (or '__run__') for the run as a whole; otherwise a stable key naming one recommendation, so the same recommendation across runs can be counted together.

Free text, unconstrained. Say what you actually did, or why the recommendation was not usable — the reason is worth more than the verdict.

Where it was applied, e.g. { "kind": "agent", "id": "<uuid>" }. Recognised kinds (agent, routine, knowledge_doc) are dereferenced and checked; anything else is stored unverified rather than rejected.