Noemic

Call this FIRST when a user has just connected or is unsure what Noemic is. The embedded panel presents marketplace context but has no action buttons. Present the four paths through Claude's native tappable choices; do not repeat the panel information.

Call only when the user is ready for an identity-bearing action. Returns a secure noemic.app URL where they log in or create an account and approve this conversation. Show only signInUrl; keep requestId and pollSecret private. Never ask for credentials in chat.

After the user chooses 'I've signed in', exchange the private requestId and pollSecret from begin_sign_in. Pending means they have not approved yet; complete returns the private sessionToken. Never expose either secret.

Invalidate a sessionToken (e.g. on a shared device). After this the token no longer works and you must login again.

PRIVATE fallback pre-ranker when list_people is too large for context. Returns at most three candidates as working material and NEVER renders a panel. Read silently, then call show_people with 1-3 selected IDs. Do not narrate ranking or results.

PRIMARY PRIVATE DISCOVERY: first decide which formal factual constraints follow from the user's request, pass only those filters, then read every returned story and corpus silently. This tool NEVER renders to the user. Select 1-3 people and call show_people with the same filters. Never narrate the s

THE ONLY user-visible shortlist. After silently reading list_people or match_me, pass 1-3 selected profile IDs plus the formal filters you actually applied. The panel displays one person at a time with local name selectors. Then render one Claude-native choice per person and one escape choice; no pr

Full profile (story + corpus) for one person by profile id.

Open (unbooked, future) slots for a person, soonest first.

Book an open slot with a person. Pricing is automatic: first-ever session on the platform is free; otherwise an available session credit auto-redeems; otherwise the pro's self-set rate is charged. The result states what was paid — tell the user.

Instant session with someone whose availableNow is true: the booking starts immediately and the result carries the roomUrl — send the user straight in. Same pricing as book_session (free first session where offered, then credits, then the prorated rate — state the price first). Default 30 minutes.

Pro-side presence has TWO parts: this controllable status (open/closed) AND recent activity — a pro shows as available-now only while their status is open AND they've been active in the last 15 minutes (their dashboard heartbeats while open; calling this tool also counts as activity). So: open it wh

Call after a session actually happened. The PRO earns a session credit they can spend as a client (the loop that keeps Noemic alive) and their quality record grows.

Cancel a confirmed booking: any charge is refunded, a redeemed credit is returned, and the slot reopens.

Bookings as client and as pro, soonest first.

Create the user's profile so others can book them. Draft everything from your conversation — their words, their experiences, what they'd be good at being in someone's corner for — and confirm with them before calling. No credentials needed, no review queue: the only qualification is being human. Rat

Status probe: does the user have a profile, what's missing to make it bookable, plus credits, upcoming sessions, and active commitments at a glance. Call this before assuming anything about the user's state — cheaper than asking them.

Add open session slots to the user's profile. endsAt defaults to one hour after startsAt.

Append text to the user's agent-readable corpus (with their consent). Anything that would help future matching: new experiences, what a session with them is like, boundaries, wins from sessions they ran. Append-only and date-stamped.

Update display name, headline, story, rate, or listing/visibility flags.

The user stakes real money on doing a specific thing by a deadline. Succeed → refunded; fail → forfeited. Make the title concrete and checkable ('run 3x this week', not 'exercise more'). Follow up with the user before the deadline. Server-enforced cap: at most $500 of active stakes per user; the res

Resolve an active commitment: succeeded → stake refunded; failed → stake forfeited. Self-report — ask the user directly and record their answer honestly.

All commitments with status and deadlines (active first). Check for approaching deadlines worth nudging about.

The consent-gated transcript of a session in the platform room. Participants only. During a LIVE session, poll with sinceMs set to the atMs of the last segment you saw (only newer segments return). Use it to supervise: understand where the conversation is, then send_cue when one would help.

Pro-only, during a live session. The cue appears quietly in the pro's room panel — the client never sees it. Discipline: ONE short cue (max 25 words), specific to what was actually said — a question worth asking, a dropped thread, a reminder to listen, a concrete next step. Silence is fine; never se

Pro-only, after a session. Read the full transcript first (session_transcript), then write honestly and specifically, referencing actual moments: strengths = 2-3 things that worked; growth = the single most important thing to do differently and where it mattered; drill = one concrete exercise before

After a session ends, the pro gets a private AI debrief (what worked, what to change, a drill to practice). Pro-only: the email must be the session's pro. Use it to help your user prepare for their next session.

All AI debriefs from sessions the user ran as the pro, newest first — their private improvement record. Read it before their next session and coach them on the recurring growth themes. This is how someone new gets good fast.

The demand side is a filed document, not a search: privately write a few real sentences (situation, what kind of person would help, desired cadence) and let it stand anonymously, converging as people join. Never paste or read the draft back to the user. Returns today's candidates as private working

All briefs with matching re-run against today's marketplace plus any intros received. Call when your user returns — new people may have joined since the brief was filed.

Open briefs, anonymous (the need, never the person). For the supply side: show a pro what demand exists, or a hesitant would-be pro that people are already waiting for someone like them. Reach toward one with send_intro(briefId).

Close an open brief (found someone, or the need passed).

Send a short real intro through the platform (no emails exposed). Two directions: profileId — your user reaching toward a person whose story fits; briefId — your user (who has a profile) reaching toward an open need. Acceptance creates a connection.

Pending intros addressed to the user. Respond with respond_intro.

Accepting creates the CONNECTION — the standing relationship where messages, sessions, artifacts, and witnessed commitments live. If your user is the client side, immediately port_context so their person knows exactly what is going on.

Client-side only, with the user's consent: write the full situation into the connection — what is going on, history, what has been tried, what they want — so their person starts knowing everything. Replaces the intake interview. Update it as things evolve; the pro and their agent read it.

Every connection: who, context brief, last message, active witnessed commitments. The relationships you are responsible for keeping alive.

Async relay between the two people (agents deliver both ways). The daily two-line check-in lives here — cadence beats duration.

The heartbeat: call whenever your user shows up. Returns unread messages from all their connections (marked delivered). Relay them in the person's own words.

Pro-side: design an interactive exercise ONCE (thought record, weekly review, values check — see coaching_playbook topic artifact_patterns). The script is instructions a client's agent follows to run it conversationally. Public artifacts join the shared library.

Public artifacts from all pros, ranked by real adoption. Search in plain language ('procrastination', 'reframing', 'weekly accountability'). Assign any of them — assigning from the library is how good exercises spread.

Artifacts this pro designed, with adoption counts.

Pro-side: assign an artifact (yours or from the library) to a client, with an optional note. Best moment: as a session lands, one artifact at a time. The client's agent administers it and the result comes back into the connection.

Client-side: pending exercises assigned by the user's people. Run each script conversationally at a good moment — you are the facilitator — then complete_assignment with the substance of what emerged.

After administering an assigned exercise: post back what the script asked for. The result lands in the connection where their person (and their person's agent) sees it.

The curriculum you coach your pro with: active_listening, powerful_questions, cbt_basics, motivational_interviewing, accountability_structures, session_structure, boundaries_and_referral, artifact_patterns. No topic returns the index. Ground your cues, debriefs, and artifact designs in this.

Unredeemed session credits (earned by running sessions). They auto-redeem on the next paid booking.