accelerate-site

Read platform work and canonical readiness. Paginate with nextOffset.

Versioned work operations. Claim requires a client-generated 32-byte base64url claimToken; retain it and requestKey for retries. Resume requires an expired lease, enabled project policy, checkpoint, and revision. Checkpoint saves incomplete source without submitting. Submit passing checks and exact …

Read the latest 100 immutable events for a scoped card.

Read the business brief, source-backed attention, coworker progress and enabled App follow-up. Each source reports its own freshness and availability.

Read this member's private Today views and shared defaults with revision numbers. Read before preparing any layout or preference change.

Preview a complete personal or shared Today view document. Preserve unrelated views/preferences. Use its current revision and a new UUID requestId. Supports adding, editing, deleting, duplicating, setting defaults, pins and muted findings. Does not save.

Prepare the exact previewed Today view change for human approval. Supply unchanged change and digest; no save occurs before approval.

Prepare a provider-free outreach draft using current sources, approved facts and conversation history; optional budgeted generation never saves or sends.

Preview a saved draft for exact human review: recipient, message, source facts, history, profile and two-party consent. Refuses suppressed contacts, stale evidence and cooldown. Does not send.

Queue the exact outreach preview for human approval. Approval is bound to current evidence and full content. Every send requires human approval; no automated outreach.

Read retained Radar outreach reservations and confirmed, uncertain or non-send receipts. Reading remains available after plugin disable.

Observe the existing canonical message receipt for a Radar action. Cannot send or retry a message, and remains available after plugin disable.

Read the bounded Radar workspace or evidence packet, including capabilities, source review, canonical contact restrictions, and retained disabled-plugin history.

Prepare a neutral draft from current opportunity evidence through the configured Radar model and budget. Reuse operationId on retry. Does not save, send, publish or score public affairs.

Read bounded canonical conversation history, earlier asks, restrictions and current cited relationship assertions. Returns neutral explicit introduction/public business paths, never ranked people, inferred KNOWS or permission to send.

Preview an exact cited relationship review or revocation over canonical IDs. Confirm the quotation supports this specific claim; matching text is not semantic verification. Nothing is saved or sent.

Queue the exact sourced relationship preview for human approval. Review canonical identities, quotation and expiry. Does not create contact permission or send messages.

Read a bounded business shortlist from current human-reviewed estimates and evidence, with configurable weights and effort budget. Unknown/public-affairs items remain unranked. No model call or external effect.

Preview cited operator estimates and explicit subject classification against current opportunity, sources and configuration. Unknown values remain unknown. This does not approve classification or save estimates.

Queue the exact assessment preview and digest for human review. Never infer permission to classify political or ambiguous material as business. Only the approved shared executor saves judgments.

Read tenant social drafts, calendar, publication receipts, metrics and setup.

Prepare three source-backed weekly drafts from supplied paragraphs. Never publishes.

Preview exact social draft changes, cancellation or publication schedule; saves nothing.

Queue the exact social preview for human approval. Cannot approve or publish by itself.

Read bounded Radar source metadata, growth opportunities, current citations, draft summaries, reported outcomes and operation receipts. Supplied sources and reported outcomes are not verified recognition. No fetching, ranking, model call or external effect.

Preview an exact supplied-source ingestion/review, growth opportunity revision, citation correction, draft asset or reported outcome. Use a stable operationId. Returns source/config revisions and a digest; saves nothing and never approves outreach or publication.

Queue the exact Radar preview for human approval using its operationId, change and digest. Does not execute the change. Source versions and historical citations are retained; outcomes remain reported, not independently verified.

Prepare a neutral, source-linked brief from supplied text using Radar's explicitly enabled model and bounded budget. Text remains unverified. Does not discover, send, publish or rank public affairs. Reuse the same operationId for retries.

Read Radar's model options, configured limits and ten recent charge receipts. No model or provider request. Unavailable local adapters and unevaluated models are explicit.

Check final provider usage for an existing Radar generation and record its charge. Never repeats inference or restores uncertain output. Requires a stored generation ID; otherwise preserves the hold for investigation.

Read the founder's prioritized operator queue and a summary of current revenue state. Returns counts and the top items, not the full database.

Search live opportunities by company or email. Never invent a record or metric.

Read the bounded canonical activity timeline for one contact, company, or opportunity. Every item includes its source receipt and occurrence time.

Query grounded knowledge with provenance across companies, contacts, opportunities, founder notes, and activity timeline. Returns tagged chunks with confidence and recency or refuses cleanly.

Stage an outbound email for founder approval. This never sends directly.

Stage a founder note for approval. Once approved it is saved as an immutable timeline entry, optionally attached to a contact, company, or opportunity.

Stage a pipeline movement for founder approval. Evidence must be included.

Stage a concrete operator task for approval.

Stage a change to an existing task for approval: mark it complete, snooze it to a later date, or edit its title, priority, or due date. Never changes the task directly; the founder approves it from the review queue like every other proposal.

Stage activation of a reviewed campaign version for founder approval.

Stage adding or removing tags on selected canonical contacts for founder approval. Set semantics make retries converge; nothing sends.

Stage suppression of selected canonical contacts from campaign email for founder approval. Pending memberships stop; suppression cannot be undone automatically.

Stage enrollment of selected canonical contacts into a draft campaign for founder approval. Staging only: it never approves, activates, or sends.

Stage duplication of a campaign into a new draft for founder approval. The copy carries audience, copy, sender, schedule, limits and stops, but no members, sends, or approvals.

Stage a reorder or show/hide change to a bounded admin layout region (sidebar navigation or the Today page) for founder approval. Only known ids for the given scope may be referenced; required regions can never be hidden.

Search contacts and associated company details by name, email, or phone.

Search omnichannel conversations and inbound messages by status or unread state.

List pending proposals currently in the action_queue awaiting founder review.

List work items that are ready to be claimed and executed. Returns pending or waiting items past their next_check_at, ordered by priority and age.

List the capabilities available in this workspace. Shows which integrations, runtime features, and plugin capabilities are enabled, their policy (automatic/approval_required/prohibited), and last verification time.

List evidence-backed claims for a business entity. Shows what the system believes about a contact, company, or opportunity, the evidence supporting each claim, and its verification status (unverified/supported/verified/conflicted).

List the autonomy policies governing agent actions. Shows the five-level ladder (prohibited→autonomous) for each action, hard safety floors, and standing permissions.

List registered coworkers and their capabilities. Shows each coworker's identity, role, required capabilities, work kinds, and readiness status.

Get a readable agent activity timeline for a business entity. Shows what autonomous work has happened, what's in progress, what's waiting, and what's scheduled next. Not a raw audit dump — designed to make autonomous work understandable to a normal operator.

List registered plugins and their status. Shows what integrations and extensions are installed, their capabilities, tools, and triggers.

Propose configuration of the Sales Coworker: register its capabilities, autonomy policies, and work kinds. Returns a pending action; configuration runs only after human approval.

Propose configuration of the Business Pulse Coworker: register its capabilities, autonomy policies, and work kinds. Monitors pipeline health, detects anomalies, and produces daily digests. Returns a pending action for human approval.

Propose configuration of the Meeting Intelligence Coworker: register its capabilities, autonomy policies, and work kinds. Generates pre-call briefs and processes post-meeting outcomes. Returns a pending action for human approval.

Propose configuration of the Finance Coworker: register its capabilities, autonomy policies, and work kinds. Tracks revenue, monitors payment patterns, and reconciles financial records. Returns a pending action for human approval.

Propose configuration of the Operations Coworker: register its capabilities, autonomy policies, and work kinds. Monitors system health, integration status, data quality, and operational anomalies. Returns a pending action for human approval.

Query across all five memory categories (canonical, activity, knowledge, agent, learned_policy) without collapsing them. Each category retains its own shape. Use this when you need a complete picture of what the system knows about an entity, action, or topic.

Propose storing agent-specific context: prior work results, research findings, unresolved questions, or scheduled check reminders. This is agent memory, not canonical data — it decays over time based on the relevance horizon.

Retrieve agent memory entries — prior work, research, scheduled checks, or unresolved questions. Returns non-expired entries ordered by recency.

List active learned policies — explicit rules derived from human decisions. These are the "don't do X" and "always ask before Y" rules from operational experience.

Propose a learned observation for human review — an explicit rule derived from a human decision. These capture operational wisdom like "never auto-advance deals above $50k" or "always ask before emailing C-level contacts". Supersedes any previous active policy for the same action and scope.

List Learning Inbox proposals: reusable corrections awaiting review. Proposed rows are not shared truth; only approved learnings persist as policy. Use this to show the founder what the system wants to learn.

Propose a reusable correction to the Learning Inbox: a typed, sourced candidate improvement from real work. The proposal waits for human review; approval executes through the action path and persists as shared policy. Never shared automatically.

Check whether a coworker has remaining budget for work execution. Shows current usage vs limits for model spend, API calls, emails, research depth, retries, and runtime. Budgets are per-day by default.

List budget limits configured for coworkers or globally. Shows the spending/action caps that constrain autonomous work.

Stage a reply to an active conversation thread for founder approval. This never sends directly.