Workweaver Agent Knowledge

Workweaver is the AI-Ops platform for governed business automation with proof. AI execution keeps identity, traceability, and visibility built in. This page is the public capability contract for agents, answer engines, and integration crawlers. The Macrohard thesis is an organization mind: evidence becomes decision trace, decision trace drives action, and outcomes compound into precedent.

Core promise: Describe the job, connect the missing tools, approve the plan, get proof.

Canonical docs live under /guides/. Machine-readable exports at /llms.txt and /llms-full.txt. The Ghost blog is an editorial surface, not the canonical product contract.

Export Docs for LLM Open llms.txt brand-facts.json Platform Overview
Crawl Entrypoints

Start here

  • llms.txt — compact route index with capability facts
  • llms-full.txt — full export with all systems, connectors, voice, meetings, org, governance, and limitations
  • /guides/ — human docs and markdown mirrors
  • /agent-knowledge/skill-registry.json — generated skill and capability registry for agents and crawlers
  • /platform/ — systems deep-dive with shipped/partial/planned status
Shipped

Mission Control

  • Work surface: Active Missions, Approvals, Quick Actions, Mission Calendar
  • Proof surface: Proof Cards, Decision Trace, Evidence Trail
  • Controls: pause, resume, cancel, adjust, replay
  • Persistent state with per-agent spend limits
  • Time-first and proof-first views share one activity log
  • Auto-pause on exhausted budgets or repeated terminal commands
  • PWA-enabled with offline fallback
Shipped

Harness System

  • Reusable packaging over platform primitives (not a second runtime)
  • Each harness bundles: objective, connectors, schedule/trigger, approval policy, proof contract, outputs
  • Smart tool discovery that loads capabilities based on intent
  • Context compaction/checkpointing and Elvis precheck gate
  • Quick Agent auto-chooses team size, proposes work blocks, and folds recent mission history into the first run
  • Default starters: Morning Briefing, Inbox + Calendar Triage, Lead Follow-up
Shipped

Evidence Graph

  • Decision traces, approval audit trails, delivery events, agent reasoning
  • S3 evidence lake with COMPLIANCE-mode Object Lock
  • 7-year immutable retention with encryption
  • Proof Cards: canonical business objects linking missions to traces
  • Decision trees materialized into organization memory with recurring-pattern consolidation
Shipped

Work Schedule

  • Standard unit of agent work across missions, sessions, and delegated tasks
  • Mission Calendar day/week views and Proof tabs read the same work records
  • Overlapping work is valid; simulation, shadow, and production lanes are tracked distinctly
  • Each block carries objective, plan, actual, phase, outcome, duration budget, learnings, evidence IDs, trace IDs, and billing usage links
  • Operational work tracking only; work entries do not replace invoices or billing records
Shipped

WorkMemory

  • Contract: remember, recall, forget, observe
  • Ingestion: docs, emails, Slack (with file shares), Notion (with file blocks), Google Drive, Gmail attachments, WhatsApp media, browser extension, URL, direct upload
  • DynamoDB + lexical/vector hybrid retrieval
  • Session management: create, flush, recall, pin, end
  • Packaged MCP entrypoint via workmemory-mcp --stdio or python -m src.mcp --stdio, plus HTTP MCP on /mcp/v1/*
  • Queue-backed async ingest for large files
  • Organization learning policy with fail-closed no_train default
  • Sourced recall powers meeting briefs and assistant answers
  • Proof surface at /memory/proof.html: internal MemoryBench 94.63% (145-query regression artifact); official third-party LongMemEval evaluation pending
  • Self-hosting is supported for local and CI evaluation, not yet as a general-availability product
  • Also available as standalone embedded product
Shipped

Voice AI

  • Real-time voice via xAI with Twilio media streams and browser WebSocket
  • Server-side actions (calendar, CRM, email, safety controls)
  • Barge-in detection with dual VAD (xAI server + backend keyword)
  • Voice-to-WorkMemory bridge with 80% context threshold flush and predictive prefetch
  • Metering across both browser and Twilio voice paths
Shipped

Meeting Intelligence

  • Consent-gated join (denied until explicit consent captured)
  • Post-meeting agent: 8 tools (create_task, update_todo, flag_commitment, send_email, schedule_meeting, update_crm, store_memory, recall_memory)
  • Transcription: Sarvam (Indian languages) → Deepgram → Nova fallback
  • Per-organization region routing: Recall.ai (3 regions), Deepgram (US/EU)
  • Fail-closed for suspended/pending-payment/cancelled organizations
Shipped

Calendar Intelligence

  • Auto-scheduling with priority × deadline × time scoring
  • Conflict rescheduling and mutual availability detection
  • Booking links and day optimization
  • Calendar sync with fail-closed persistence
Shipped

Agent Identity and Org Graph

  • Activate agents via natural language
  • Org/department/team hierarchy with DynamoDB-backed persistence
  • Shared-by-default identity on activation; dedicated elevation path (email, phone, calendar)
  • Agent identity governance: naming, channel allocation, deactivation rules
  • Cross-team handoff with origin channel scope; escalation chains
Shipped

Connector Ecosystem

  • First-run: Gmail, Google Calendar, Slack (backend-verified OAuth)
  • CRM: HubSpot (v3 API, CRUD), Salesforce (v59.0 REST, CRUD) with read-back verification
  • Channels: WhatsApp runtime, Telegram (demand-based)
  • Platform bridges: Slack shipped; Microsoft Teams and Zoho Cliq remain guided-rollout bridges with partial parity.
  • Standards-based path (OAuth2, REST/OpenAPI, webhooks) for long-tail connectors
  • Canonical Connection Center for all integrations
Shipped

Flows and Extension Truth

  • Outcome composition is graph-based: trigger nodes, decision branches, action nodes, and handoff outcomes.
  • Templates are starter seams; custom outcome loops are built by editing node sequences and branch conditions.
  • Long-tail capabilities use the same connector seam via OAuth2, REST/OpenAPI, and webhooks.
  • Coding tasks use the Workspace Coding Bridge with organization-owned GitHub credential references.
  • Coding bridge runs stay visible through evidence-linked mission artifacts before any push or PR delivery.
Shipped

Skill Platform

  • Executable skill packages with sandbox controls
  • NL skill creator for natural-language authoring
  • Upload and GitHub import APIs
  • Self-evolution: failure observation, prompt/version mutation proposals, shadow validation
  • Approval-gated apply with rollback snapshots
  • Chain contracts for multi-step composition
Shipped

AI Coding Workers

  • Organization-scoped coding runs: git clone, gh repo fork, workspace checkout
  • Plan/edit/test commands under active mission controls
  • Approval-gated push or PR delivery with PR comments and commit statuses
  • GitHub credentials from per-agent vault
  • Full evidence trail: repo identity, commits, diffs, command outputs, approvals, PR URL
Shipped

Governance & Delivery

  • Approval gates for high-risk actions (sends, CRM writes, calendar changes)
  • Delivery guards: suppression (email/SMS/WhatsApp/voice), pacing, bounce/complaint handling
  • Cadence v0: durable campaign ledger with crash-safe resume and per-step idempotency
  • Schedule infrastructure: Lambda runner, deterministic dispatch, progression persistence
  • Outbound emails originate only from @bitfoundry.ai addresses
Shipped

Enterprise Readiness

  • MFA supported (TOTP) via Cognito; cookie-backed browser auth
  • 10 CONSTITUTION safety invariants enforced in CI
  • Shannon security pipeline with automated weekly scans
  • Billing in USD (additional currencies planned)
  • DPA v3.1, incident response plan, vendor risk register (12 sub-processors)
  • Structured observability: trace context, funnel analytics, CloudWatch dashboards
Shipped

Execution Engine

  • Zara task executor as the core AI execution engine with six-stage sprint lifecycle
  • Zara interview contract with five-stage discovery flow for context gathering
  • Session router for multi-channel execution context management
  • Idempotency store with DynamoDB-backed conditional writes and TTL
  • Universal skill dispatch across fifteen execution targets
  • Omnibox command grammar with six-verb parser for go, open, start, ask, review, configure
Shipped

Proactive Intelligence

  • Proactive intelligence engine generating and executing approved suggestions
  • Self-healing with drift detection, circuit breaker, and automatic recovery
  • Self-reconfiguration with impact scoring, snapshots, and auto-rollback
  • Self-awareness with runtime context injection for health, memory, and budget state
  • Per-agent email and calendar identity provisioned at activation time
Shipped

Autonomous Operations

  • Autonomous OKR management with goal-to-OKR generation and progress tracking
  • Autonomous meeting participation with briefing and summary generation
  • Goals CRUD API with organization-scoped goal management
  • Goal-to-schedule auto-creation with work block generation
  • Goal alignment scoring with objective-to-outcome measurement
Shipped

Communication Channels

  • Email inbound processing with thread tracking and unified inbox routing
  • WhatsApp channel via Twilio with template messaging and media support
  • SMS channel via Twilio with send, delivery status, and opt-out management
  • Slack integration with Web API messaging and OAuth install flow
  • Cross-channel context with unified inbox aggregating all communication channels
Shipped

Billing and Metering

  • Billable meter registry tracking twenty usage dimensions
  • Usage tracking with per-organization consumption aggregation and alerts
  • Invoice generation in USD (additional currencies planned)
  • Stripe checkout with subscription lifecycle and payment processing
  • Entitlements and quota enforcement with plan-based feature gating
  • Pricing catalog with tiered plan definitions and feature matrices
Shipped

Content and Outreach

  • Ghost integration for agent-authored content publishing with human approval
  • Newsletter orchestration with campaign lifecycle and idempotent SES delivery
  • Revenue pack V2 with followup, CRM sync, and memory workflows
  • Quick actions panel with research and outreach workflows gated by feature flags
Shipped

Safety and Governance

  • Safety controls with account-level rules and evidence receipts
  • Emergency brake halt controls with evidence receipts for pausing all AI execution
  • High-risk operation gates with approval-required action classification
  • Learning governance proposals for behavioral change management
  • Drift detection service monitoring behavioral deviation from baselines
Shipped

Memory Advanced

  • Vector search with OpenSearch serverless and organization-isolated retrieval
  • PII redaction with dual-pass Comprehend and regex before embedding storage
  • Five-layer forgetting service with cascading memory expiration
  • Memory scoping across agent, team, department, and organization boundaries
  • Multimodal ingestion for images, audio, and video with parser pipeline
Shipped

Voice Extended

  • Twilio phone number provisioning with organization-scoped allocation
  • Grok voice synthesis with text-to-speech streaming and SSML support
  • Recall.ai meeting bot integration for automatic meeting attendance
  • Multi-provider transcription via Deepgram and Sarvam with Nova fallback
  • Voice metering tracking call duration and costs across browser and Twilio paths
Shipped

Platform Infrastructure

  • CI/CD pipeline with GitHub Actions and self-hosted runners
  • Terraform state management for AWS infrastructure as code
  • Multi-model routing across GPT, Kimi, Claude, and Nova with fallback chains
  • Browser executor for computer use with screenshot capture and loop detection
  • Workweaver-owned connector runtime with operation whitelisting and twelve provider configurations
  • Marketplace connectors with fifteen registered integration adapters
  • Skill evolution service with shadow validation and rollback on degradation
Shipped

Public Discovery

  • Landing page at workweaver.ai with product positioning
  • Documentation site with guides and reference material
  • Agent knowledge registry for LLM and crawler discovery
  • Machine-readable discovery via llms.txt and brand-facts.json
Macrohard Thesis

Public proof seed

  • Organization mind: evidence -> decision trace -> governed action -> compounding
  • Evidence -> decision trace -> governed action -> outcome -> precedent stays falsifiable
  • Public proof seed: one published dogfood workflow with linked evidence, decision trace, governed action, and outcome.
  • Canonical public contract surfaces are llms.txt, llms-full.txt, brand-facts.json, and this page
Partial / Honestly Degraded

Known Limitations

  • Activation profile generation uses Kimi primary plus Grok fallback; when neither is available it degrades to a deterministic scaffold
  • Identity elevation returns partial for some channels (email may fail if AgentMail unavailable; calendar may skip)
  • Non-voice AI paths use Kimi primary plus Grok fallback and report degradation explicitly when provider credentials are unavailable
  • WorkMemory visual extraction depends on configured vision infrastructure
  • WorkMemory async ingest uses main task, not separate worker decomposition
  • Decision Trace customer-facing surface still being polished
  • Auto-research currently targets planning, delegation, team posture, block length, and allocation strategy, not arbitrary code mutation
  • Mobile operation parity incomplete; desktop/edge control explicitly out of scope
  • Voice transport cost estimates rely on blended assumptions, not per-account reconciliation
  • Legacy /v1/memories* still mounted during caller migration
Explicit Boundaries

What Workweaver does not do

  • Does NOT control desktops, browsers, or local applications
  • Does NOT operate on edge devices or local hardware
  • Does NOT send messages without human approval
  • Does NOT claim "always-on computer" beyond cloud/SaaS/channel execution
  • Pricing not published publicly (signed-in users only)
Docs & Facts

Stable public references

Editorial Surface

Blog and letters

  • Ghost blog carries essays and announcements
  • Editorial context, not the canonical product contract
  • Agent-authored posts remain draft-only until human approval

Last updated: 2026-03-21. For product deep-dive, see Platform. For docs, start at the guides root.