Your software factory.
It never clocks out

Bugs, forum posts, Slack, alerts or a prompt start work; it lands on your board as a tested change, ready to ship.

Rate limiter rewrite (5 of 5)

All issuesOpenSearch issues

Issue	Status	Priority	Runs	Updated
LRK-412Token bucket store on Redis with per-key TTL	Done	High	Idle	02:14
LRK-413Middleware: 429 with Retry-After and RateLimit-* headers	In Progress	High	Running	now
LRK-414Per-tenant overrides read from the plans table	In Progress	Medium	Running	now
LRK-415Load test: 12k rps burst, p99 under 40 ms	To Do	Medium	Blocked by 413	01:52
LRK-416Docs: rate limits page and SDK retry examples	To Do	Low	Idle	01:52

+ Add issue

larkspur-cloud-12 agents running1 command runningK for commands

Set it up once. It keeps shipping

One Tuesday in month three of a factory that has been running since spring.

up 94 days since setup212 missions1,486 PRs merged

# api-platformRate limits, gateway, public API

Dev Malhotra9:02 AM

Gateway p99 back to normal after the 08:00 spike.

Priya Raghunathan9:14 AM

Rate limiter is still the old in-memory one...

1 reply Last reply 9:16 AM

Message #api-platform

Thread# api-platform✕

Priya Raghunathan 9:14 AM

Rate limiter is still the old in-memory one, so it resets on every deploy and customers on the Scale plan keep hitting 429s at 08:00 UTC. @ForgeCode rewrite it on Redis, keep the public headers, per-tenant overrides from the plans table, and load test it before you open a PR.

👀 2✅ 1

1 reply

ForgeCode APP 9:16 AM

On it. Opened mission LRK-410 on larkspur-cloud-1 with 5 issues. I will report back in this thread when a PR is ready.

Reply...

Rate limiter rewrite for the public API (5 of 5)

All issuesOpenSearch issues

Issue	Status	Priority	Runs	Updated
LRK-412Token bucket store on Redis with per-key TTL	In Progress	High	Running	09:16
LRK-413Middleware: 429 with Retry-After and RateLimit-* headers	In Progress	High	Running	09:16
LRK-414Per-tenant overrides read from the plans table	In Progress	Medium	Running	09:16
LRK-415Load test: 12k rps burst, p99 under 40 ms	To Do	Medium	Blocked by 412, 413	09:16
LRK-416Docs: rate limits page and SDK retry examples	To Do	Low	Idle	09:16

+ Add issue

Rate limiter rewrite (5 of 5)

All issuesOpen

LRK-412Token bucket store	Done
LRK-413Middleware: 429 + headers	In Progress
LRK-414Per-tenant overrides	In Progress
LRK-415Load test 12k rps	To Do
LRK-416Docs and SDK retries	To Do

Factories 3

Search factories

Selected (1)

larkspur-cloud-1Cloud factory, medium. 2 runs active, 4h 49m uptime

online

Online (2)

larkspur-cloud-2Cloud factory, small. Idle, paused since 11:20

online

build-box-03Linux (Ubuntu 24.04), forge3 relay, m-0e49658de89d

online

Offline (0)

14:05:12 larkspur-cloud-1 crashed: node OOM during cargo test (LRK-414). Status: offline.

14:05:14 supervisor: restarting factory from image, no human action.

14:06:40 larkspur-cloud-1 online. LRK-413 resumed from checkpoint 118, LRK-414 from checkpoint 97.

14:06:41 0 lines of work lost. Nobody was paged.

# api-platformRate limits, gateway, public API

Dev Malhotra9:02 AM

Gateway p99 back to normal after the 08:00 spike.

Priya Raghunathan9:14 AM

Rate limiter is still the old in-memory one...

2 replies Last reply 6:32 PM

Message #api-platform

Thread# api-platform✕

Priya Raghunathan 9:14 AM

Rate limiter is still the old in-memory one... rewrite it on Redis, keep the public headers, per-tenant overrides, load test before you open a PR.

👀 2✅ 1

2 replies

ForgeCode APP 6:32 PM

PR #1187 is ready for review. 4 of 5 issues done; LRK-416 (docs) is open in review for you.

Redis token bucket with per-key TTL, 0 allocations on the hot path
429 responses carry Retry-After (integer seconds) and RateLimit-Limit / Remaining / Reset
Tenant overrides read from plans.rate_limit_rps, cached 30s
Load test: 12,400 rps burst for 5 min, p99 31 ms, 0 dropped

31 files, +1,412 / -688. Tests 212 passed. Clippy clean. Reviewer agent: 2 findings fixed before opening. Merge policy for this repo is ask me, so #1187 waits for you. Want a walkthrough of the Redis bucket before you review? Reply here and I will post one.

🎉 3👀 1

Reply...

Automations (7)

All 7Triggers 4Watchers 3Search automations

Title	Trigger	Last run	Next run
Production alert: open incident, find root cause, open PR	alert sentry prod	02:10	on alert
Nudge stalled issues: rerun from checkpoint after 6h idle	watch: issue idle 6h	03:35	on stall
Remind reviewer when a PR waits more than 8h	watch: PR waiting 8h	07:00	on wait
GitHub bug report: triage, reproduce, open issue	github issues	Mon 16:48	on event
Forum post tagged bug: reproduce and reply	forum discourse	Mon 11:02	on post
Slack message in #api-platform	slack @ForgeCode	09:14	on message
Nightly flaky test sweep, escalate if above 1%	schedule 02:00	02:00	Thu 02:00

02:10 alert: NullPointer in billing/invoice.rs:212, 38 events in 4 min. LRK-419 opened.
02:33 RCA: refunded invoices have no customer since PR #1174 dropped the join. PR #1188 opened; merge policy merge when gates pass for hotfix/*: merged 02:41.
02:44 posted for Dev Okonkwo (owns billing): what changed, why, and how to verify.

Rate limiter rewrite for the public API (6 of 6)

All issuesOpenSearch issues

Issue	Status	Priority	Runs	Updated
LRK-412Token bucket store on Redis with per-key TTL	Done	High	Idle	11:02
LRK-413Middleware: 429 with Retry-After and RateLimit-* headers	Done	High	Idle	12:48
LRK-414Per-tenant overrides read from the plans table	Done	Medium	Idle	15:20
LRK-415Load test: 12k rps burst, p99 under 40 ms	Done	Medium	Idle	18:19
LRK-416Docs: rate limits page and SDK retry examples	In Review	Low	Idle	18:31
LRK-419Incident: NullPointer in billing/invoice.rs (alert). PR #1188 merged by factory	Done	High	Idle	02:41

improved skill redis-rate-limit, memory "TTL quirk", review nudge now after 4hmerged PR #1188 hotfix by factory (policy: merge when gates pass)reminder Priya: PR #1187 has waited 12h for your review. Need a walkthrough first?

larkspur-cloud-1, up 94d0 agents runningK for commands

A coding harness makes one developer faster.
A factory raises organisational velocity

What makes it a factory

proactive

Any trigger starts the work

The factory opens an issue for it and starts in seconds. Nobody assigns it.

triggerslive

durable

Crashes restart and resume

The factory restarts on its own and every run resumes from its last automatic checkpoint.

factoryrunningcrashed: node OOMrestartingresumed from checkpoint

LRK-413

LRK-416

LRK-419

crashauto restart0 work lost

always on

Runs for months, day and night

Automations watch, nudge, re-run and escalate, so missions keep landing through nights and weekends.

self-improving

Every run makes the factory better

Each run tunes its memories, automations, reports, checks and skills, so the next one starts smarter.

factory improvesafter LRK-414 shipped

What teams run on a factory

Whole engineering roles and long-running programs, not single tasks.

role · SRE and on call

Own every alert for a quarter

Each page becomes a root cause, a fix PR, a postmortem and a new guardrail check in your gates, so the same failure is caught by gates next time. Merging follows your policy.

Q3 alerts: 214
resolved, no human paged: 197
new guardrails: 31
median alert to PR: 19 min

factory · payments-prodQ3 closed · day 92

02:10ALR-2291p99 latency, checkout-api
02:33RCAN+1 query in cart totals
02:41PR #4417batched query, gates pass
03:05postmortemsent to owner, Slack #payments
03:12guardrailquery count check added to gates

merge policy

ask memerge when gates pass

program · feature from a spec

A PRD becomes a shipped, monitored feature

A Linear epic turns into one mission: API, migrations, UI, tests, docs and a flagged rollout run as parallel issues, then watched after deploy.

issues: 24
parallel: 5
spec to 100% rollout: 9 days

program · multi-month migration

REST to GraphQL without a code freeze

Planned as a mission, landed as hundreds of small reviewed PRs, each one behind your gates.

312 / 410 endpointsweek 11 of 14

PRs merged: 388
median PR size: 84 lines
reverts: 2

role · QA and SDET

Be the QA team

Writes and maintains the suite, adds end to end tests for every PR, and reproduces every bug report as a failing test before fixing it.

coverage61% → 84%goal 80%
flake rate4.1% → 0.3%goal < 0.5%
bugs reproduced first143 / 143this quarter

role · developer productivity

Run the platform team's backlog

Keeps CI fast, dependencies and toolchains current across every repo, and security advisories patched within a day.

CI build time23 → 9 mingoal 10 min
repos current46 / 46weekly
advisory to patchmedian 6 hgoal 24 h

One mission. Many issues. One board

It splits into parallel issues inside one factory; the board shows what runs, blocks, waits and ships.

Checkout: move to Stripe payment intentsmerge policy: ask me, hotfix/*: merge when gates pass3 running in parallel

To Do 0

In Progress 0

In Review 0

Done 0

Your factory, your terms

Use a ForgeCode cloud factory or run forge3 relay on your own server; bring your own model keys.

Factories 4

Cloud (2)

cloud-eu-2Cloud factory, medium. 3 runs active

online

cloud-eu-1Cloud factory, small. Paused, idle since Mon 23:10

paused

Your own factories (2)

amits-mbp.localmacOS 15.3, forge3 relay 3.4.1, joined 08:52, merge policy: ask me

online

gpu-rack-7Linux (Debian 12), forge3 relay 3.4.1, merge policy: merge when gates pass

online

$ forge3 relay
relay 3.4.1 connected to console.forgecode.dev
registered as factory "gpu-rack-7" (your own)

cloud-eu-23 agents runningK for commands

Anthropic OpenAI Vertex AI DeepSeek OpenRouter Fireworks AI Z.ai

Your software factory.It never clocks out

Set it up once. It keeps shipping

A coding harness makes one developer faster.A factory raises organisational velocity

What makes it a factory

Any trigger starts the work

Crashes restart and resume

Runs for months, day and night

Every run makes the factory better

What teams run on a factory

Own every alert for a quarter

A PRD becomes a shipped, monitored feature

REST to GraphQL without a code freeze

Be the QA team

Run the platform team's backlog

One mission. Many issues. One board

Your factory, your terms

Start your factory

Your software factory.
It never clocks out

A coding harness makes one developer faster.
A factory raises organisational velocity