Specdevrel.md/0.2
ProductAcme Vector
URLhttps://example.com/acme-vector
Stagegrowth
UpdatedMon Sep 28 2026 00:00:00 GMT+0000 (Coordinated Universal Time)
Owner[email protected]
DEVREL.md
Fictional example. Acme Vector is not a real product. The numbers show the format, not real benchmarks.
Product
Acme Vector adds vector search to an existing Postgres database as an extension plus a hosted service. It's for backend and platform engineers who want semantic search without running a separate vector store. Category: vector databases and search infrastructure.
Value proposition
Add semantic search to an existing Postgres app in one afternoon, without running a separate vector cluster.
ICPs
Teams adding semantic search to an existing SaaS app
- Technical context: Python or TypeScript services, Postgres on AWS RDS or Aurora, Terraform. Acme Vector plugs in as an extension on the Postgres they already run, plus a hosted indexing service
- Company stage and team size: Series A to C, 5 to 30 engineers
- Use case: in-app semantic search and retrieval for an AI feature, on data they already hold
- Hiring moment: the product team commits to an AI search feature on the roadmap, and the first prototype needs embeddings next to existing rows
- Goals: 1. ship search without new infrastructure, 2. keep customer data in one database for compliance, 3. predictable cost as data grows
- Pains: 1. sync jobs between Postgres and a separate vector store, 2. on-call for another cluster, 3. weeks lost comparing vector databases
- Alternatives: a dedicated vector database, left when the sync jobs and second cluster cost more than they return; pgvector run in-house, left when index tuning eats platform time; doing nothing and keeping keyword search, left when users expect results by meaning
- Switching forces: push: sync jobs break and results go stale. Pull: one database, with the backups and permissions they already have. Anxiety: query latency at scale, and extension support on RDS. Habit: the team already knows its current search setup
- Decision: a staff engineer evaluates and champions it; the VP Engineering approves spend
- Activation event: first production query served from an existing table, within 7 days of signup
- Fit score: 3.8 (Pain 4, Urgency 4, Activation ease 3, Strategic value 4)
Startups building a chat or agent feature on their own data
- Technical context: TypeScript, Next.js, Vercel, Supabase or Neon. Acme Vector plugs in through the TypeScript SDK against their hosted Postgres
- Company stage and team size: seed, 2 to 8 engineers
- Use case: retrieval for a chat or agent feature
- Hiring moment: answers from their agent are wrong because it cannot see the customer's data
- Goals: 1. get retrieval working this week, 2. stay on the free tier until there is revenue
- Pains: 1. too many retrieval tutorials that don't match their stack, 2. tuning chunk sizes by trial and error
- Alternatives: pgvector on Supabase or Neon directly, left when they need ranking and filtering they would otherwise build; a managed vector database's free tier, left when they want one fewer service
- Switching forces: push: retrieval quality is hurting demos. Pull: works with the Postgres they already have. Anxiety: another vendor to pay later. Habit: copying the retrieval code from a tutorial that already half works
- Decision: the engineer adopts alone, card payment
- Activation event: first query from a deployed app, within 48 hours
- Fit score: 2.9 (Pain 3, Urgency 4, Activation ease 4, Strategic value 2)
Anti-personas
- Solo builders and hobby projects: high signups, no budget, no path to expansion
- Enterprises that require on-premise deployment with custom SLAs
- Teams not on Postgres
North Star
- North Star metric: weekly active projects with a production query, 120 today, 250 target (product analytics, Sep 2026)
- Time to Hello World: 22 min median today, 5 min target
- Measured: signup to first successful similarity query, product analytics, last 30 days
- First success means: a query returns ranked results from a sample table the developer loaded
Activation
- Event: first production query against an existing table (not the sample data)
- Target: within 7 days of signup
- Current rate: 14% of signups within 7 days, about 48 activated developers a month
- Retention: unknown. The window will be 30 days, since production search is used daily
Funnel health
| Stage | Gate | Now | Pass |
|---|---|---|---|
| Awareness | > 50% of signups fit an ICP segment | 58% of September signups fit a segment, from enriched signups tagged by profile and first-week usage | yes |
| Onboarding | Median time to first call < 5 min and first-call success > 80% | 22 min, 64% | no |
| Activation | Activation rate > 20% and > 60% of activated developers active after 30 days | 14%, retention not yet measured | no |
| Engagement | Community answers > 65% of questions within 24h | unknown | unknown |
| Monetization | Paying deepens trust rather than replacing it | Too early to judge | n/a |
Metrics
| Stage | Leading | Now | Lagging | Now |
|---|---|---|---|---|
| Awareness | Docs to quickstart click-through | 0.8% | Signups per month | 340 |
| Onboarding | Quickstart completion | 38% | Median time to first query | 22 min |
| Activation | Activated in 24h | 6% | Activated developers a month | 48 |
Docs map
- Quickstart: https://example.com/acme-vector/docs/quickstart
- API reference: https://example.com/acme-vector/docs/api
- SDKs: Python, TypeScript
- llms.txt: https://example.com/llms.txt
- MCP server: none
Voice and guardrails
- Plain and technical. No "revolutionary", no "AI-native".
- Never claim benchmark results without linking the method.
- Security and compliance guidance is written or reviewed by a human.
Open questions
- Where do developers drop off inside the quickstart? Loading sample data is suspected.
- Does anyone answer community questions systematically?
- How many activated developers are still querying after 30 days? Needed for the Activation gate.