Persisted leader/subagent ownership. An explicit ping resumes an independent peer; a managed session stays under its team rule.
Persisted leader/subagent ownership. An explicit ping resumes an independent peer; a managed session stays under its team rule.
(ask env question)(ask env question opts)As a subagent, ask your parent for a decision and wait for the reply text. Returns None when the parent is not running or does not reply within timeout_secs (default 600). Then choose the safest option and say so in your answer.
As a subagent, ask your parent for a decision and wait for the reply text. Returns None when the parent is not running or does not reply within timeout_secs (default 600). Then choose the safest option and say so in your answer.
(cancel-agent env session-id)Cancel a subagent and its descendants, including queued work. Returns its record.
Cancel a subagent and its descendants, including queued work. Returns its record.
(checkpoint! env messages)Capture the last complete model input, before its in-flight tool call. Never copy runtime handles.
Capture the last complete model input, before its in-flight tool call. Never copy runtime handles.
(claim-iteration! env)Reserve one model iteration. A child budget is durable across replies, turns and restarts.
Reserve one model iteration. A child budget is durable across replies, turns and restarts.
(controls? db actor-id target-id)Leaders control their descendants. Subagents control themselves and their direct children.
Leaders control their descendants. Subagents control themselves and their direct children.
(enabled? db sid)Whether session sid can use managed subagents. A new session starts without the session
settings of its source, so a managed child reads the setting of its leader. Without such a
session setting, or with an invocation value, the configuration decides.
Whether session `sid` can use managed subagents. A new session starts without the session settings of its source, so a managed child reads the setting of its leader. Without such a session setting, or with an invocation value, the configuration decides.
(reconcile-orphans! db live-sids)Fail the children that a dead process left queued or running. Spare a child or a leader
that live-sids names: a live process still owns that team. Returns the count.
Fail the children that a dead process left queued or running. Spare a child or a leader that `live-sids` names: a live process still owns that team. Returns the count.
(require-enabled! db sid)Refuse new managed work before bootstrapping a session or consuming a checkpoint.
Refuse new managed work before bootstrapping a session or consuming a checkpoint.
(restrict-router env router)Apply an inherited allowlist to every request and retry, including automatic fallback.
Apply an inherited allowlist to every request and retry, including automatic fallback.
(routing-change env before)The route command issued for this session since before, as
{:preference {:provider :model} :command command}, or nil. Agent routing and the
engine's authentication rescue issue commands; a manual pick is composer state that
the next submitted turn snapshots, so it never re-routes a running turn.
The route command issued for this session since `before`, as
`{:preference {:provider :model} :command command}`, or nil. Agent routing and the
engine's authentication rescue issue commands; a manual pick is composer state that
the next submitted turn snapshots, so it never re-routes a running turn.(run env tasks)(run env tasks opts)Run each task in its own subagent and wait for all of them. A task is a string or a dict with task and optional model, provider, iteration_budget and context. A child uses your model and your full context. A child on another model gets only the system prompt and its task, unless context is "full". Children share the checkout.
Records have session_id, status, answer and question. More than 16 tasks start as earlier ones finish. timeout_secs defaults to 240 (max 1800). If a status is still running, call council.wait. lost means the child stopped without an answer.
Run each task in its own subagent and wait for all of them. A task is a string or a dict with task and optional model, provider, iteration_budget and context. A child uses your model and your full context. A child on another model gets only the system prompt and its task, unless context is "full". Children share the checkout. Records have session_id, status, answer and question. More than 16 tasks start as earlier ones finish. timeout_secs defaults to 240 (max 1800). If a status is still running, call council.wait. lost means the child stopped without an answer.
(stop-reason env)Give the cause when claim-iteration! refuses the child session of env.
Give the cause when `claim-iteration!` refuses the child session of `env`.
(wait-agents env)(wait-agents env ids-or-opts)(wait-agents env ids opts)Wait again for subagents by session_ids (default: all your children). Same records and timeout as council.run.
Wait again for subagents by session_ids (default: all your children). Same records and timeout as council.run.
(wake-allowed? db author-id recipient-id)Automatic resume has two sources: an explicit ping between independent sessions, and a managed team. A child may return to its leader; siblings must share a task team. A managed session keeps its team rule, so a cancelled or exhausted subagent stays idle. Only a managed subagent wakes itself; an independent session stays idle.
Automatic resume has two sources: an explicit ping between independent sessions, and a managed team. A child may return to its leader; siblings must share a task team. A managed session keeps its team rule, so a cancelled or exhausted subagent stays idle. Only a managed subagent wakes itself; an independent session stays idle.
cljdoc builds & hosts documentation for Clojure/Script libraries
| Ctrl+k | Jump to recent docs |
| ← | Move to previous article |
| → | Move to next article |
| Ctrl+/ | Jump to the search field |