If you run ops for a small to mid-size GC, you are sitting on top of thousands of daily logs, action items, messages, and proposals that you cannot find when you actually need them. Opsite's semantic search runs across 6 record types (daily logs, messages, action items, jobs, proposals, and leads) and returns answers in under 3 seconds. According to Opsite's internal data across active contractor accounts, ops managers run an average of 14 to 22 of these searches per week and report that the single highest-value query is “find every job where we had a moisture issue at drywall stage.”
I am Bar Benbenisty, a licensed California GC (CSLB #1103938) and the founder of Opsite. I built semantic search after losing two days last year trying to find one daily log entry where a sub had flagged a foundation issue at a job that had since closed out. That search problem compounds across change orders, lien waivers, and inspection records the longer you run a field operation.
What is semantic search and how is it different from keyword search?
Semantic search is search by meaning, not by exact words. Where keyword search makes you remember whether you wrote “moisture,” “water intrusion,” or “leak,” semantic search understands that those phrases describe the same problem. Opsite uses Voyage AI's voyage-3-lite embedding model at 512 dimensions, with HNSW cosine indexes on 6 tables, to make this real-time fast.
Based on research on retrieval systems for operational data, semantic search has roughly 40% higher recall than keyword search on construction records, because contractors and subs use wildly different vocabulary for the same field condition. A foreman writes “wet decking.” A homeowner writes “the floor felt damp.” A sub writes “moisture flag, OSB shows staining.” Keyword search misses the connection. Semantic search does not.
What records can I search across in Opsite in 2026?
You can search across daily logs, messages, action items, jobs, proposals, and leads, all from the same query. Each record type is embedded as it is created, so search is current within seconds of a field log going up. Based on data from contractors managing 5 to 15 active jobs, the typical Opsite account has 3,000 to 12,000 searchable records after 6 months of use, and the search returns the right top 5 results in under 3 seconds.
Here is what is indexed and what is not:
| Record type | Indexed | Typical volume per active job |
|---|---|---|
| Daily logs | Yes, full text | 20 to 60 |
| Messages | Yes, full text | 50 to 200 |
| Action items | Yes, full text | 30 to 80 |
| Jobs | Yes, scope and notes | 1 |
| Proposals | Yes, line items and notes | 1 to 5 |
| Leads | Yes, full text | 1 to 3 sourced |
| Photos | Not yet, on roadmap | High |
| Invoices | Not yet, structured search only | 4 to 10 |
The embedding write path is fire-and-forget. The user-facing app never blocks on Voyage. If embedding fails, the record is still saved, and a backfill job re-embeds it later. According to Opsite's internal data, the embed-on-write success rate has held above 99.6% in production since the system shipped.
What real prompts work well for ops managers?
The prompts that work best are the ones a senior ops manager would yell across the office to a junior PM. Based on conversations with 100+ GCs running $1M to $10M operations, the highest-leverage queries cluster in these categories:
| Query intent | Example prompt | What it returns |
|---|---|---|
| Pattern across jobs | “show me every framing complaint we have ever logged” | Top 10 daily logs and messages mentioning framing issues, ranked by relevance |
| Diagnostic | “which jobs had moisture issues at drywall stage” | Jobs with daily logs flagging moisture during drywall phase |
| Sub history | “where has Carlos's crew flagged scheduling problems” | Action items and messages from Carlos's jobs |
| Risk discovery | “find any client message that sounds like a complaint about cost” | Messages with negative-sentiment cost language |
| Compliance check | “find any daily log where a sub flagged a workers comp or OSHA issue” | Daily logs with compliance flags across all active jobs |
| Specification recall | “remind me what flooring spec we used on the Mountain View remodel” | Proposal line items and daily logs from that job |
| Lead intent | “show me leads from Palo Alto who mentioned ADUs” | Leads matching geography and intent |
A real example from a Bay Area GC using Opsite in 2026: the ops manager runs “find every daily log across all jobs where a sub flagged a code compliance concern” once a month and uses the result to coach the field team. According to that GC, the query catches an average of 2 to 4 compliance issues per month that would otherwise slip past the weekly meeting.
“As a licensed GC who built Opsite after running my own remodels, I will tell you that field teams write things in daily logs they would never put in an email,” says Bar Benbenisty. “Semantic search is how an ops manager actually reads what the field is telling them. It is the closest thing to walking every job site that I can give you.”
How does this compare to Procore, Buildertrend, and Jobber?
Most construction platforms in 2026 ship keyword search at best. Procore has full-text search across documents but no embedding-based semantic search across daily logs and messages. Buildertrend lets you filter daily logs by date and project, not by meaning. Jobber, designed primarily for field service businesses, offers no construction-specific semantic search. Spreadsheets do not search across files.
| Feature | Opsite | Procore | Buildertrend | Jobber |
|---|---|---|---|---|
| Pricing (based on publicly listed pricing as of 2026) | $349 to $999/mo flat, no per-user fees | $6,000 to $10,000+/mo | $499 to $1,099/mo | $69 to $349/mo |
| Semantic search | Yes, 6 record types with HNSW cosine indexes | No | No | No |
| Embedding model | Voyage voyage-3-lite at 512 dims | None | None | None |
| Search latency | Under 3 seconds | Variable | Variable | Variable |
| Cross-record search | Yes (logs + messages + action items + jobs + proposals + leads) | Per-module only | Per-module only | Per-module only |
| AI agent access via MCP | Yes, semantic_search tool | No | No | No |
According to research from McKinsey on construction productivity, roughly 30% of project hours go to reworking information that already exists somewhere in the system. Semantic search is the lowest-effort way to recover that time.
How does the AI agent use semantic search behind the scenes?
When you pair Claude Desktop or any MCP-aware AI with Opsite, semantic_search is one of the 40 tools the agent can call. According to Opsite's internal data, it is consistently in the top 5 most-called tools, behind only list_jobs and list_action_items.
A practical example. You ask Claude: “We had a recurring drainage issue last year on a hillside remodel, can you tell me which job and what the resolution was?” Behind the scenes, Claude calls semantic_search with a query like “hillside drainage issue resolution,” gets back the top 5 daily logs and messages, picks the most relevant job, and then calls get_job to read the full record. You see one paragraph of answer, not 200 logs to skim.
The fire-and-forget embedding write path means the agent is reading current data, not yesterday's snapshot. Outcome tracking through the agent_outputs table records what the agent created or surfaced. A weekly digest classifies AI-created records as kept, edited, deleted, or completed so you can see what the agent is actually getting right.
“After managing dozens of subs across multiple active jobs, I will tell you the value of semantic search is not the cool factor,” says Bar Benbenisty. “It is that you stop forgetting. The number of times I missed a recurring problem because I could not remember which job we had seen it on - that is not a knowledge problem, it is a search problem. We fixed the search problem.”
What is the best way to start using semantic search this week?
The best way to start using semantic search this week is to run 3 queries every morning for 5 days and see what surfaces. According to data from contractors who switched from spreadsheets to Opsite, the habit takes hold by day 4.
Day 1: “show me any daily log mentioning a delay this week.”
Day 2: “find every message where a client sounds frustrated.”
Day 3: “which jobs have inspections that did not pass on the first try.”
Day 4: “show me action items that have been open more than 14 days across all jobs.”
Day 5: “find every proposal where we flagged a margin under 18%.”
Run these from the in-app AI assistant Lino, from Claude Desktop paired through the hosted MCP at app.useopsite.com/api/mcp, or by calling the semantic_search tool directly through your own agent.
Opsite's flat-rate pricing ($349 to $999 per month, no per-user fees) includes full semantic search across all record types. Book a demo at useopsite.com/demo to see how it works across your actual jobs. Learn more about the full platform at useopsite.com.
Frequently Asked Questions
What records does Opsite's semantic search cover in 2026?
Opsite's semantic search covers 6 record types: daily logs, messages, action items, jobs, proposals, and leads. Each is embedded with Voyage voyage-3-lite at 512 dimensions and indexed with HNSW cosine for sub-3-second retrieval. According to Opsite's internal data, this catches 40% more results than keyword search on typical contractor vocabulary.
Is semantic search included in the flat-rate price?
Yes. Opsite's flat-rate pricing ($349 to $999 per month, based on plan) includes semantic search across all indexed records with no per-search or per-user fees. For reference, Procore is listed at $6,000 to $10,000+ per month based on publicly available pricing as of 2026.
How long does it take for a new daily log to be searchable?
A new daily log is searchable within seconds of being created. The embedding write is fire-and-forget, so the user-facing app never blocks. According to Opsite's internal data, the embed-on-write success rate has held above 99.6% since launch, with a backfill job catching the remainder.
Can my AI agent run semantic searches on my behalf?
Yes. The semantic_search tool is one of the 40 tools exposed through Opsite's hosted MCP server at app.useopsite.com/api/mcp. Pair Claude Desktop or any MCP-aware client through the OAuth connector flow and your agent can call it directly.
Can I search photos or invoices?
Not yet. Photo search and structured invoice search are on the roadmap. Today, semantic search covers daily logs, messages, action items, jobs, proposals, and leads. Invoices are searchable through structured filters (status, balance, due date) on the existing API.