Built for software teams.
Stop being the bottleneck for every cross-team question.
Pulse is the company-brain compiler. It connects to your tools, builds a permission-aware map of how work actually flows, and compiles your team’s repeating procedures into portable SKILL.md files any AI agent can execute. Every employee gets a homepage that knows their context.
Pulse builds a live map of your company: people, decisions, docs, customers — and how they actually connect.
Three structural choices, all interlocked. Each is hostile to the economics of an enterprise tool, which is what makes Pulse work for software teams of 5–500.
Pulse models features, decisions, and commitments as first-class entities, not just documents. That's what enables Feature Archeology, Decision Memory, and accurate effort estimation.
A 30-minute interview with the founder builds a working model of your processes, vocabulary, and pain points. Connectors come second. No 4-week IT-led setup.
$19/seat base, $5 per verified agent action. Aligned to value created, not headcount. Hostile to the per-seat model that prices Pulse out of SMBs.
Every recurring procedure (refund handling, incident response, deal sign-off) becomes a portable SKILL.md file. Drop the bundle into a git repo. Load it in Claude Desktop, Cursor, or any Skills-compatible client. Pulse ships a compiler, not just an index — your process graph becomes executable.
Daily cron embeds new entities and scores divergence over 14 days. Skills past 0.6 flip to DRIFTED + write a notification. The export warns AI clients that the procedure has aged.
Every invocation logs an outcome inferred from the related AgentAction's status. Action policies can require minSkillConfidence ≥ 0.85 before a skill-proposed action runs auto vs manual.
Same skill, different SKILL.md per role. Pulse infers per-step required teams from the source entities and binary-redacts steps the viewer can't see. Compiled output respects every ACL the source data already has.
Cosine > 0.8 between skills surfaces conflicts (keep-separate, archive-one, defer). DRAFT → SHADOW → ACTIVE lifecycle: shadow skills observe humans on similar tasks; ≥85% match auto-graduates.
--- name: rollout_pricing_change description: How to roll out a new pricing tier across docs, billing, and customer comms. triggers: - "we're changing pricing" - "ship a new pricing tier" - "deprecate a price point" version: 1.0.0 calibration: invocations: 12 success_rate: 0.92 --- ## 1. Confirm the decision is logged in Decision Memory Pull the entry. If the rationale isn't crisp, ask the decision-maker for one sentence — every downstream artifact will cite it. ## 2. Update pricing-doc-v3 in Notion Replace the tier table; preserve the prior version under `archive/`. Tag #pricing-strategy in the change log. ## 3. Wire the new tiers into Stripe Create the new Stripe Products. Mirror the old tier IDs in `metadata.legacy_tier_id` so analytics doesn't break.
Five working destinations plus an AI agent layer that drafts the Slack DM, the Linear ticket, the calendar invite, and waits for one click. Click any item in any surface and you flow into the next: Home → Living Context → Feature Archeology → Decision Memory → Ask → Search → Planner → Agents. Home is the daily dashboard you start from; the destinations below are where the work happens.
Confidence scores, reasoning trails, conflict surfacing, request-access flow. Answers that admit when sources disagree, when an answer is stale, and when something is hidden behind permissions.
Cross-entity keyword search over people, decisions, docs, customers, features, topics, playbooks. Intent-aware grouping puts what you meant first; pinned canonical answers float to the top. ResourceAcl filtering so private channels stay private.
Decision Memory: the structured archive of what was decided, by whom, with what rationale. Feature Archeology: full timelines of every project. Feature Planner: similar past features inform realistic estimates.
Living Context Cards that auto-update from underlying data. Expert Finder ranked by behavior, not credentials. Role inheritance briefings when someone takes over.
Describe what you're scoping. Pulse pulls similar past features, gives you an effort range from your team's actuals, generates a pre-mortem, suggests a team, maps stakeholders, and drafts a starter doc.
Every surface threads into a calm action layer. Pulse drafts the Slack DM, the Linear ticket, the calendar invite — lands them in your approval inbox. One click sends. 5-minute undo on every external write. No autonomous fan-out, ever.
Most enterprise-search products are built around a doc graph and a vendor-level trust contract. Pulse picks a different model on each axis. Three places the divergence shows up:
Doc graph: pages, chunks, embeddings.
Process graph: decisions, commitments, features as first-class entities. Built into the schema, not bolted on.
Search-time ACL filter. Index everything; hope the filter doesn't leak.
ACL-mirrored at retrieval. Native source-system permissions enforced. Request-access flow when restricted content is needed.
Vendor-level trust: the company is SOC 2, therefore the answer is too.
Sentence-level trust: every passage marked cited or inferred, every retrieval shows the chain, every doc carries a quality score and a decay curve.
Trust signals at the sentence level, memory that ages content gracefully, AI that argues back, process visibility that frames the system instead of the person, and an action layer that lets AI do what it would otherwise just suggest, with your approval, and a compiler that turns your team’s repeating work into portable SKILL.md procedures any AI agent can execute. 110 capabilities across ten clusters; here is a curated set.
AI doesn't just suggest. Drafted DMs, Linear tickets, calendar invites land in your inbox; one click sends them. 5-min undo on every external write.
see in productEvery Ask answer exposes the 5-stage chain (query, retrieval, weighting, rank, synthesis) with boost/demote signals shown inline.
see in productEach decision gets an outcome-check timestamp. Pulse re-evaluates at the right moment and surfaces 'this aged badly' so the lesson lands.
see in productAuto-updating cards for every person, customer, project. Hover any name in any surface, get the full picture. Embedded in Pulse, Slack, and Chrome.
see in productPulse-themed keyword search over people, decisions, docs, customers, features, topics, playbooks. Intent-aware grouping puts what you meant first; pinned canonical answers float to the top.
see in productPulse watches the process graph and compiles repeating procedures (refund handling, incident response, hiring loops) into Anthropic Agent Skills format. Loadable in Claude Desktop, Cursor, and any Skills-compatible client.
see in productPulse is the surface. The five tools are the substrate.
Three things changed in her areas overnight. Her direct report Marcus is OOO. A decision from yesterday's all-hands was extracted and surfaced for confirmation. Pre-meeting briefing for her 9:30 1:1 with Therese: what Therese's working on, last 1:1 commitments, recent blockers. Ninety seconds. Aligned.
Asks Pulse: 'Has anyone debugged this here before?' 73% confident answer in 2.4 seconds. Two past Slack threads, one fix PR, three ranked people. Pulse surfaces a conflict (older guidance vs. newer) and tells her which is current. She pings Yusuf, who already has her context because Pulse showed him the Living Context Card.
Opens Map → Feature Planner. Pulse finds four similar past features. Excel upload took 3 weeks (encoding edge cases). Bulk user import 5 weeks (validation rules longer than expected). Pulse generates a starter doc with realistic estimates and pre-built risks. Adekunle edits, saves to Notion. Forty minutes saved.
Asks Ask in deep-research mode: 'Map last quarter's eng output: features shipped, time-to-ship vs. estimates, blockers, velocity.' Ninety seconds. Structured report: 12 features, estimate-vs-actual variance, the auth dependency that delayed three things, 15% velocity gain post-onboarding. Confidence 81%, full citation trails. Four hours saved.
Three drafts waiting. Therese's PR has been stuck 18 hours, so Pulse drafted a Slack DM to Vikram, the reviewer. Two 1:1s skipped this week, so a Calendar invite for Friday. A decision from Monday's standup without a captured rationale, so a Linear comment asking the owner. Clarissa tweaks one tone, hits approve on all three. They send. A five-minute undo button shows on each, then quietly disappears. Three follow-throughs that used to fall on the floor.
Home shows three things: a promise he made yesterday is approaching its window, a teammate's question in #design has gone two hours unanswered (Kabir is the most likely person to know), his PR has been waiting on review for eighteen hours. He sends, answers, pings. Closes Pulse.
Two paths into every AI client. Today: download a SKILL.md tarball from /api/skills/export and load it in Claude Desktop, Cursor, or any Skills-compatible tool. Next: Pulse-as-MCP at mcp.pulse.appfor live queries — same permission scope, no bundle to refresh. Both routes mirror the requesting user’s ACLs.
Drop the Pulse Skills bundle into Claude Desktop and Claude knows how your team handles refunds, incidents, deal sign-off — with the rationale and provenance attached.
Your IDE inherits the architecture decisions, the failure cases, and the procedures the team has codified. Compiled per-user so private context stays scoped.
The SKILL.md export is portable today; the MCP server (5 tools: search, decisions, experts, recent activity, your Pulse) is the next ship.
If you can't see it in Slack, you can't see it in Pulse. Never expand access.
Pre-decisional or HR-adjacent? The AI helps you think but doesn't remember.
Mark anything 'forget' and it's purged from the index. Auditable but gone.
Every employee sees exactly what Pulse knows about them. Exportable.
We’re onboarding teams from the waitlist. Tell us about yours and we’ll be in touch.