Every feature that earns its keep. Nothing that doesn’t.
110capabilities across the five product surfaces, the trust layer, the agent-actions layer, and workspace operations. Each one earned its slot by solving a problem real teams flagged in interviews, not by checking a competitor’s box. Every card links to the live screen.
Daily context, calmly delivered.
What changed since you looked, what's stuck, what's coming. Pulse surfaces signals on a homepage you check, never as push notifications. The opposite of an inbox.
Diff mode
What changed since you last looked, scoped to your work. Not a feed; a curated delta.
Promise tracking
Commitments you've made or are owed, surfaced as they approach due dates. Pulled from threads, not entered manually.
Pre-meeting briefings
Two hours before a meeting, Pulse compiles last-1:1 commitments, recent blockers, and current PRs to ground the conversation.
OOO awareness · stuck detection
Pulse knows who's out, what their work blocks, and which items have been stuck on you for 18+ hours. Always you-owns or blocking-you, never both.
Workload signal (you-only)
A personal pulse on your in-flight commitments. Not surveillance, never a manager-visible metric. Designed to be ignored when busy.
Daily suggestion engine
One prioritized nudge per day from your context: what to do, why now, why it's the highest-leverage move on your plate.
Pattern interruption alerts
When you've broken your own cadence (a 1:1 missed for 14 days, a doc you usually open weekly untouched), Pulse calls it.
Reverse delegation
Pulse spots when someone keeps asking you the same question. Usually it's permissions, not knowledge. Get them what they need once.
Cross-timezone handoff briefings
Your APAC counterpart goes online in 2h. Here's what changed today, what they need to weigh in on, async-ready digest.
Stakeholder fatigue detection
Pulse warns when you're over-loading someone ('12 messages to Apoorv this week vs typical 5'). Saves relationships before they fray.
True meeting cost
Every meeting gets a $-cost badge based on attendee comp bands and length. Makes the recurring 7-VP standup visible before it's been running for a year.
Learning moments capture
When old work suddenly gets cited 8× in a week, Pulse nudges you to refresh it before the next round of citers misses the update.
Conversational Q&A across your work.
Confidence scores, reasoning trails, conflict surfacing, request-access flow, and a memory that follows the thread. Answers that admit when sources disagree, when an answer is stale, and when something is hidden behind permissions.
Multi-turn conversation memory
Follow-ups understand the prior turn. 'What about cancellations?' after 'how does annual billing work?' just works. Topic shifts auto-start a new thread.
Entity cards in Ask answers
Type 'Tell me about Therese' or 'status of Project Athena' and Pulse renders a card strip above the answer: people, customers, features, decisions, topics. Click through to the entity's detail page.
Brainstorm mode
Fourth Ask mode: divergent, no verdict, no calibration. Just 5-7 distinct ideas. Different from Counter-evidence (adversarial); brainstorm is for opening up the option space.
Inline jargon hover
Hover over team-specific terms in any Ask answer (ICP, RICE, project codenames) and Pulse explains using context from your workspace. New hires onboard 2× faster.
Take action from inside Ask
While reading an answer, click 'Take action' and Pulse drafts a Linear ticket / Slack DM / Notion page from it. Lands in your /app/actions inbox for approval.
Calibrated confidence + reasoning
Every answer carries a 0-100 confidence score plus an expandable reasoning trail. When Pulse says 87%, it's right ~87% of the time on this workspace.
Conflict surfacing
When sources disagree, Pulse shows both views with the recency-and-authority hierarchy that resolves them. Never silently picks one.
Why am I seeing this?
Every result is explainable. One click reveals which sources got retrieved, which got weighted, and why this answer over alternatives.
Hallucination self-flagging
Sentences are tagged 'from sources' (lime underline) or 'inferred' (iris dotted) with hover-to-see reasoning. Honest by default.
Counter-evidence mode
Flip the question. 'I want to renew Vendor X. What's the case AGAINST?' Pulse searches for disconfirming evidence, weighted highest first.
Behaviour-based source weighting
Sources weighted by who actually uses them. A doc opened weekly by the decision-maker beats one no one's touched in 6 months.
Provenance chains
Every Ask answer exposes a 5-stage chain (query → retrieval → weighting → rank → synthesis) with boost/demote signals shown inline.
Question history per topic
Topics asked >3× this month get surfaced. Pulse promotes a canonical answer or flags a doc-update need.
Knowledge decay curves
Pricing decays in 30 days. Mission decays in 5 years. Pulse weights freshness in retrieval per category half-life.
Devil's advocate mode
Draft a decision. Pulse pulls 4 pushbacks from your own org's history: past failure cases, conflicting decisions, current signals.
Adjacent possible explorer
What companies in your shape have tried. Curated corpus of case studies, S-1s, founder posts, ranked by relevance to you.
Time-shifted Q&A
Ask 'what would I have known on date X?' Built for retros, decision audits, and onboarding context.
Constraint relaxation thinking
What changes if budget were unlimited? Time? Headcount? Pulse runs each relaxation with your actual context and shows the tradeoffs.
Decision-to-execution gap
Every decision links to the PRs and issues meant to execute it. Pulse computes the lag from decided to merged and flags decisions that quietly never got built. No more 'we decided that 6 months ago, what happened?'
Disagreement archive
When sustained back-and-forth resolves, Pulse captures the structure: position A, position B, evidence cited, who resolved it, rationale. Stops the same call from getting re-litigated 6 months later.
Cross-entity keyword search
Pulse-themed search at /app/search across people, decisions, docs, customers, features, topics, playbooks. Intent-aware grouping (typing 'marcus' surfaces People; 'pricing' surfaces the pinned canonical answer) with ResourceAcl filtering so private channels stay private.
Saved-search alerts (social listening)
Tap the bell on any query and Pulse re-checks every 6 hours, diffs against the last run, drops a notification on new matches. First-run is a silent baseline so subscribing doesn't spam you. Mute or remove from the same surface; admin oversight at /app/admin/saved-searches.
How you got here, and the shape of work.
Most products give you a knowledge graph (docs and people). Pulse gives you a process graph: decisions, commitments, features, and bottlenecks as first-class entities. The structural bet baked into our schema.
Decision Memory
Structured archive of decisions extracted from Slack, meetings, docs. Browseable by project, person, topic. Each entry has rationale, status, and outcome.
Counterfactual replay
What if we had decided differently? Pulse simulates the alternative timeline against your actual context: pre-mortem after the fact.
Failure case library
Patterns this team has been burned by (encoding bugs in CSV imports, cross-team auth dependencies that stall, customer-churn silences). Read once, internalize.
Decision regret tracking
Each decision has an outcome-check timestamp. Pulse re-evaluates at the right moment and surfaces 'this aged badly' so the lesson lands.
Pre-mortem generation
Before you commit, Pulse asks 'imagine this fails: why?', drawing failure modes from your past projects + adjacent-possible corpus.
Process checkpoints
Recurring multi-step workflows get checkpoints. When a stage stalls, Pulse asks 'what's blocking?' instead of letting it rot.
Workflow archeology
Trace any current state back through the decisions that created it. The CSV-format choice, the auth design, the pricing model: every 'why is it this way' answerable.
Cross-tool de-duplication
Same work captured across Slack threads, Linear issues, Notion docs gets unified into one entity. Stop counting the same project three times.
Disagreement archives
When a disagreement is resolved, both positions + the resolution path get captured. Avoids re-litigating the same call six months later.
Bottleneck detection
Stages taking 2× or more vs baseline get surfaced. Always frames as the process, never the person.
Async meeting alternative
Post a question, route it to N attendees, gather responses on their schedule, AI synthesizes the agreement / disagreement / resolution.
Expertise and context, where it lives.
Living Context Cards that auto-update from underlying data. Expert Finder ranked by behavior, not credentials. Role inheritance briefings when someone takes over. Knowledge transfer is a first-class workflow, not an afterthought.
Living Context Cards
Auto-updating cards for every person, customer, project. Hover any name in any surface, get the full picture. Embedded everywhere you work in Pulse.
Skill graph
Inferred skill matrix from PR co-authoring, doc edits, meeting attendance, Slack answers. Updates without anyone filling out a form.
Expert Finder
Describe a problem, get a ranked list of people who've worked on similar things. Behavior-ranked, not credential-ranked. Includes availability state.
Manager second-brain
Per-direct-report rolling context: recent commitments, blockers, retention signals, growth conversations. Privacy boundary: never feeds metrics or rankings.
Quiet contributor surfacing
ICs whose work is invisible to the manager (code reviews, async help, doc cleanup) get attributed weight in Pulse without surveilling chat.
Role inheritance briefing
When someone takes over (promotion, transfer, replacement), Pulse compiles a comprehensive context dump: ongoing work, key contacts, in-flight items, and a week-1 triage map drawn from the predecessor's last 90 days.
Champion tracking
Living context for every customer: sentiment trend, ticket volume, champion activity, renewal timing. Internal handoff when a champion leaves.
Champion-left alerts
Daily watch fires the moment a customer's champion changes role or leaves. Notification + the existing customer.champion_changed webhook event. Renewal risk surfaced before it bites.
Departure mode
When someone gives notice, Pulse compiles their working surface into a structured handoff: channels, docs, decisions, frequent collaborators.
30/60/90 onboarding curriculum
Personalized curriculum drawn from the departing employee's collaboration graph + the team's failure cases. Hire view ⇄ manager view.
SOPs from observed work
When the same multi-step workflow happens repeatedly, Pulse drafts the SOP from the observed steps. Ship it or edit it.
Crowdsourced Q&A · karma
When Pulse can't answer, route to a likely expert. Karma + integration with daily workflow gives the structural incentive prior attempts missed.
Customer voice aggregator
Synthesizes support tickets, sales calls, NPS comments, churn interviews, and social mentions into themes per quarter. Real visibility for product teams.
Trust at the sentence level.
Most enterprise AI blurs the line between cited and generated. Pulse marks each sentence with where it came from, scores every doc, and gives every employee a control panel for what's stored and what's forgotten. Trust is the only feature that compounds.
Calibration tracking
Pulse tracks its own confidence vs accuracy per workspace + per topic. If 80%-confident answers are right 65%, the system knows and surfaces it.
Weekly digest
Friday afternoon summary of what shipped, what's stuck, what's worth knowing. Curated, not exhaustive. Opt-out, not opt-in.
Sentiment pulse (team-only)
Aggregate-only signal from Slack tone + meeting cadence. Surfaces 'something's off' to managers without naming individuals. Tunable per team.
Personal data dashboard
Every employee sees what's indexed about them, what's been inferred, every AI decision involving their data. Self-service · no admin required.
Selective amnesia · forget
Right-to-be-forgotten built in. Forget a thread, a doc, a topic, and Pulse re-synthesizes around the gap. Required for GDPR, useful past it.
Confidential mode
Per-thread, per-doc opt-out from Pulse ingestion. The lawyer-with-Slack-channel scenario doesn't compromise the rest of the system.
User correction memory
When you mark an answer wrong, Pulse remembers, not just for that question but for adjacent topics. Each user's calibration drifts toward their own truth.
Personal Graph
Tell Pulse what you own, lead, and care about. Curated beliefs (role, expertise, projects, team) feed retrieval weighting so answers tilt toward your context. Inferred beliefs are markable as wrong without losing the audit trail.
Request-access flow
When an answer would need restricted content, Pulse shows what's locked + a one-click request to the doc owner. Permissions never expand silently.
Smart staleness detection
Pulse reads the content of each doc and flags time-sensitive claims that have aged out. Goes beyond last-edited-date.
Document quality scoring
Every doc gets a 0–100 score across clarity, completeness, freshness, and citations. Re-graded weekly.
Document verification badges
Owners click 'verified accurate as of today' on a doc; a green ✓ badge appears in every Ask citation. Expires per category half-life so the badge stays honest.
Active oversight scanning
Nightly scan flags accidental oversharing: credentials in public Slack channels, revenue forecasts in tenant-wide docs, NDA-marked content with anyone-with-link share. Never auto-remediates; surfaces for admin review.
The answers your team keeps re-asking, pinned and shareable.
Knowledge isn't just stored, it's surfaced. Pulse promotes the answers your team needs most to canonical entries and gives every important doc a memorable URL. Discovery becomes a habit, not a search.
Pinned canonical answers
When a question gets asked 5+ times, Pulse prompts an expert to write the canonical answer. Future asks see it pinned at the top, verified by [author], stamped with the date, with the live synthesis still below in case facts shift.
Go Links
Short URLs like pulse.app/go/onboarding that resolve to important docs. Power-user staple. Once your team has 20 of them they start using them in Slack like punctuation.
AI that does, not just suggests.
Pulse drafts the actual action (the Slack DM, the Linear ticket, the calendar invite) and waits for one click before sending. Read-only connectors never silently start writing. Every external write is undoable. AI agency without AI roulette.
Action drafts inbox
Every AI-proposed action lands as a draft card with a per-kind preview. You see the exact recipient, exact message, exact target before anything fires. Approve / Edit / Skip.
Verified agent actions + quota meter
Every executed action counts toward your monthly quota. The marketing pricing line is now backed by a real meter, not a future promise.
5-minute undo window
After an action runs, an Undo button appears for 5 minutes. Pulse retracts the external write: Slack chat.delete, Linear comment delete, GitHub DELETE, calendar DELETE, Notion archive, Drive trash.
Outbound writes to all 8 source systems
Slack, GitHub, Linear, Calendar, Notion, Drive, Confluence, Jira. Each connector has an explicit one-time scope-upgrade consent step; read-only never silently becomes write.
Action policies
Pre-approve recurring patterns: 'DM stalled PR owners after 18h', 'Schedule retro for decisions with regret ≥ 60'. Each has a rate limit + dry-run preview that replays last week's data.
Playbooks
Bundle 3-5 actions to fire as one click. Built-ins: kickoff a feature (Linear epic + Slack post + calendar invite + Notion brief), champion-left response, decision-regret retro, onboarding day-1.
Featured playbooks
Admins pin the 2-3 most useful playbooks to a Featured section so new users discover them first instead of staring at a flat list.
One-click rollback
Every playbook edit snapshots the prior state. Bad edit? Restore last week's version with one click. Audit trail stays intact (rollback creates a new version, not a truncation).
Programmatic actions via REST API
Third-party integrations propose, list, approve, and reject agent actions via the existing public REST surface. POST /v1/actions with idempotency keys for retry-safe automation.
Cron-style playbook scheduling
Weekly digest playbook every Monday at 9am? Cron picker on the editor with presets (every-15-min, hourly, daily-9am, weekly, monthly) plus custom-cron + IANA timezone. 5-minute minimum interval.
Conversation starters on playbooks
Each playbook card shows 2-3 starter prompts. New users know exactly what the playbook does and what to type, instead of staring at a flat description.
Per-user playbook favorites
Star playbooks you use a lot. They pin to your personal 'Your pins' section above Featured. Distinct from admin-curated Featured.
Per-playbook analytics
Each playbook shows its run count, success rate, and top runners. Data-driven decisions about what's working and what's collecting dust.
Domain allowlists for actions
Per-policy guardrails: 'only DM @trellis.com employees', 'only post to these Slack channels', 'only create Linear tickets in OPS team'. Blocked actions never fire and don't burn quota.
Slash commands in decision drafts
Type /pulse in /app/decisions/draft for inline AI: /rewrite, /evidence, /counter, /summarize, /tighten, /expand. /evidence pulls top-3 retrieval hits as inspiration.
Build playbooks with AI
Type a one-line description ('alert me on critical Sentry errors in #incidents') and Sonnet drafts a working playbook with steps, payload templates, and trigger. You tweak in the editor before enabling. Auto=false on every step so humans approve.
Visual playbook canvas
Form / Visual toggle on every playbook editor. Drag between node handles to set step dependencies; cycles are blocked client-side. Same step model under both views, lossless toggle.
Pulse compiles your team's actual work into portable Skills.
Pulse ships a compiler, not just an index. Every recurring procedure (refund handling, incident response, deal sign-off) becomes a SKILL.md file, validated against drift, calibrated against outcomes, and rendered per-user with permission-aware redaction. Drop the bundle into a git repo; load it in Claude Desktop, Cursor, or any Skills-compatible client. Your process graph becomes executable — the structural differentiator.
Auto-extracted SKILL.md from your process graph
Every 12h Pulse walks Decisions clustered by Topic, FailureCases with codified `whatWeDoNow`, resolved Disagreements with procedural resolution text, and ProcessCheckpoints with stages defined. When a pattern repeats ≥3 times, Sonnet compiles it into a numbered SKILL.md with YAML frontmatter following the Anthropic Agent Skills standard.
Per-step provenance graph
Each `## N.` step carries a chip: 'cites: 1 decision · 2 failure cases'. Click any chip to drill into the source entities. Built from `SkillStepProvenance` rows so the answer to 'why does step 4 say check with finance?' is one tap away.
Drift detection nobody else ships
Daily cron embeds new entities, finds the nearest skill, scores divergence over 14 days. > 0.6 flips the skill to DRIFTED + writes a notification. < 0.3 auto-recovers. Drift score surfaces in the export frontmatter so AI clients loading a stale skill get a warning to verify with a human first.
Confidence calibration from real outcomes
Every skill invocation logs an outcome inferred from the related AgentAction's status. The nightly calibrator computes a weighted-by-recency success rate. Action policies can require `minSkillConfidence: 0.85` before a skill-proposed action runs auto vs manual. Honest routing, not vibes.
Permission-aware compilation (the structural moat)
Same skill, different SKILL.md per role. Pulse infers per-step required teams from the source entities (decision-maker's membership, FailureCase.team, document ResourceAcl) and binary-redacts steps the viewer's teams can't cover. Compiled output respects every ACL the source data already has — no reverse engineering required.
Shadow-mode canary deployments
DRAFT → SHADOW → ACTIVE lifecycle. Shadow skills don't execute; instead the harness observes humans on similar tasks and scores match-rate. ≥10 invocations + ≥85% match → auto-graduate. ≤50% kicks back to DRAFT. Canary deployments for organizational knowledge.
Conflict detection + resolution
After every compile, cosine similarity > 0.8 against existing tenant skills writes a `SkillConflict` row. Admin resolves via keep-separate (with `scopeTeam` tags), archive-one, or defer. When the team has two ways to handle refunds, Pulse surfaces the multiplicity instead of arbitrarily picking one.
Markdown + git portability (tarball export)
GET /api/skills/export?format=tarball returns a USTAR archive: README + per-skill SKILL.md, ready to drop into a git repo. Per-user compiled (your ACLs apply). Live today; live MCP server is next month.
Useful in week one. Calibrated by week two.
No procurement-first dance. No 4-week IT engagement. A 30-minute conversational interview, five OAuth connectors, and Pulse is producing answers against your real data, with permissions mirrored from the source tools, not configured separately.
Admin onboarding · 2-week PoC
Connect five tools, see the permission preview, get a calibrated proof-of-concept against your actual data within 14 days. No procurement-first dance.
Conversational onboarding
Pulse interviews each new user (role, goals, who they work with most) and uses the answers to seed their personal graph. Five minutes, replaces a survey.
MCP server · Pulse-as-MCP
Coming soonStandardized Model Context Protocol exposing Pulse's data to any MCP-compatible AI client (Claude, Cursor, ChatGPT, Cowork). Per-user permission-checked.
Meeting capture integration
Direct integration with Otter, Granola, Fireflies, Read, Zoom Cloud Recording. Meeting transcripts feed Decision Memory and Briefing mode.
Notification channel design
Eight categories (briefings, asks, stuck items, OOO, decisions, promises, meeting suggestions, access). Each tunable Slack/in-app/email/none.
Push API for custom data
Long-tail integrations without a native connector? POST /v1/documents pushes any doc into Pulse: search, Decision Memory, commitment extraction, and on-ingest sensitivity classification all wired automatically. Idempotent on retry.
Knowledge graph as a queryable layer
POST /v1/graphql gives developers a real GraphQL endpoint over the workspace graph: decisions, features, commitments, topics, people, customers, disagreements. Read-only, depth-capped, with relationship traversal. Your graph, your queries — exposed for you to build on.
Wired into every tool you already use.
The features prospects ask about second, but the ones that determine whether Pulse becomes infrastructure or becomes a tab. Connectors, BYOK, federated learning, the individual tier: Pulse fits the way procurement, finance, and contractors actually buy.
Cross-team meeting suggestion
When two teams have overlapping work but no joint touchpoint in 30 days, Pulse suggests a sync, with the participant list and proposed agenda.
Bring-your-own-LLM-key
Pipe your own Anthropic / OpenAI / Azure / Bedrock key. Procurement-friendly · cost-passthrough · audit-ready.
Open-source connectors
All nine connectors (Slack, GitHub, Notion, Linear, Calendar, Drive, Confluence, Jira, meeting transcripts) are open-source. Read the code, audit the data flow, run them yourself if you want. Trust by transparency.
Individual tier
Coming soonSolo plan for consultants, founders, IC contractors. Connect your own tools, get the full Ask + Map surface, no team required.
Federated learning · Pulse Network
Coming soonCross-tenant insights without cross-tenant data. 'Companies your size estimate features in 4w but ship in 7w', calibration from 47 customers, your data never leaves.
See it in action.
The fastest way to understand Pulse is to use it. Start a 14-day trial. No credit card, no commitment, full feature surface.