IntakeAnyone files from Claude or ChatGPT, in a real conversation.The board and DocsEvery request with its brief, size, worth and seat, and the specs and decisions beside it.Smart routingThe best seat and the best model for every request.The recordEstimate against actual, signed. Better with every row.ConnectionsClaude, ChatGPT, Claude Code, Codex, GitHub, Slack and more.
Business teamsSales, success, ops: send the customer's request and hear when it lands.Product teamsSpecs and prototypes that arrive ready to build, and a say in what runs.EngineeringThe right work, on the right seat and model. For engineers and leads.
Docs and helpEvery verb, every surface.Developerscurl -fsSL sleds.ai/install | shTrustWhat we keep, and the exceptions.ManifestoWhat we are doing, and why.Claude connectormcp.sleds.ai/mcpSelf-hostThe record in your cloud, on your keys.ContactTalk to the people building it.
Request an inviteSign in
sleds/Platform/The record
Engineering · smart routing
the record · the system of record for AI-built work

Priced before it runs. Signed when it lands.

Everything a tracker keeps, and what it never could: for every request, the price before it runs, what it was worth, who or what built it, what it actually took, and a signed receipt. Each one closes its estimate and makes the next one better.

Request an inviteThe size
QW-142auth · ssolanded · signed

SSO redirect loop on the Acme tenant

briefAcme admins loop between login and the tenant picker. Done: an admin signs in once and lands on the dashboard. Smallest: the redirect only.

  1. 2:14 pmSamantha requested it from Claude · spec.md, prototype.html attached
  2. 2:15 pmSized M · $4–11 · 20–55 min · 81% from 12 rows in auth/sso
  3. 2:15 pmRouted to Jake · Sonnet 5 · now · owns SSO, 62% headroom
  4. 2:49 pmPR #1187 merged · CI green first try · +212 −38
  5. 2:49 pm31 min · $5.20 · in band · signed ed25519 · Samantha told in her thread
requested by
Samantha · product
worth
high · Acme renewal
size
M · $4–11
seat
Jake
model
Sonnet 5
actual
31 min · $5.20
receipt
verified
a tracker keeps
  • Title and description
  • Status and assignee
  • Points someone guessed
  • A link to the PR
the record keeps all of that, plus
  • The brief, from the conversation that asked for it
  • What it is worth, and who said so
  • A price and time band before anyone starts
  • The seat and model that built it, and why
  • What it actually took, and a signed receipt
the size · a letter, a band, a confidence, a cap · never a point figure

Size it before it runs. A letter to glance at, a band behind it.

/sled estimate in Claude Code, "size this" in a conversation, or nothing at all: a spec saved to the repo and a request filed from intake are sized on their own. Story points died when the volume rose and t-shirt sizes are the same guess in fewer letters, so the size here is derived, never typed, and calibrated from your own actuals per task type, per model, per repo.

Letter, then band

M is the glance; $4–11 · 20–55 min is the truth. The letter comes from the band, so nobody argues about points in a planning meeting again.

Fit, not price

Which model fits comes from first-try CI pass, rework and parked-on-question rates for this task type in your record.

Cap

A spend cap rides on every estimate. An overnight job stops there; a session at the wall is told what is left.

Calibration

Every actual moves the next band. The share of estimates landing in band is on the roll-up, for the workspace, never per person.

the actual and the receipt · closes the estimate · calibrates the next

The receipt: what came back, who built it, signed.

When the PR merges or the overnight job lands, the row gets its receipt: minutes, model or person, cost, and a signature over the job, the outputs and the time with the workspace key. Anyone with the public key checks it locally. Which model built something is a fact, not a claim, in a compliance review or a retro. The estimate it came from closes, and the next size is a little better.

Work done on a seat carries minutes, model and seat credits; work done on the platform key carries dollars. The three rails stay named apart in the record too.

pull request #412 · body · posted by the record on merge
SLEDS-Source:    ENG-482 · spec §3 · the board
Worth:           high · $5k–20k · Acme renewal
SLEDS-Estimate:  $4–11 · 20–55 min · Fable fits · 81% · cap $20
SLEDS-Seat:      now · Fable · jake@ · seat quota
SLEDS-Actual:    34 min · $7 · fable-5.1
SLEDS-Attest:    ed25519 · 2026-09-11T14:02Z · verified
 
→ ENG-482 moved to Done · est → actual on the row
the roll-up · by area

Value in flight, landed, waiting. Cost fourth.

The same record, one zoom out. Leads see where work is piling up and what it cost.

areawhat is happeningworthcost
billing4 waiting on review · 2 days median$48k waiting—
exports2 at a wall · policy: hold$26k · resets thu 9:00—
auth6 in flight · 4 on agents$31k in flight$140 so far
platform9 landed this week · all receipted$35k landed$180
unvalued5 rows with no worth statedask the filer—
app.sleds.ai · overview · pile-up · by area · illustrative
the dataset · better with every model

What a "3-point" ticket actually cost.

Estimates against actuals, across agents, models and people, per task type, per repo. It is the dataset nobody upstream can build, it gets more valuable with every model release, and it is yours. Including the rows from our own pull requests.

slack · #eng · friday digest
This week on the sled  illustrative
 
landed      47 rows · $212k worth · $1,140 cost
in band     78% of sizes · calibration moving up
overnight   19 rows at half price · $71 saved
walls       12 hit · 12 priced · 0 waited
review      billing is the queue · 2 days median
unvalued    5 rows · ask the filer
invite only

Know what anything costs again.

Request an invite. Your first estimates close on your first merges.

We'll only use it to reach you about sleds.