Key facts
| Router model | plugsky-fusion escalates per request across tiers (live) |
| Agents | Memory, tools and orchestration are live |
| Models | 30+ models; enrichment and drafting on cheap tiers, strategy escalated |
| Tool safety | Scoped keys, RBAC and audit logs for CRM writes |
| Pricing | Flat monthly self-serve plans with no per-token charges on self-serve |
| Free tier | plugsky-micro and plugsky-lite on the free plan, no card required |
| Usage controls | Scoped keys and usage analytics per sequence or team |
| Roadmap | The batch endpoint is coming soon for overnight enrichment jobs |
TL;DR
- Enrichment, summaries and CRUD are cheap-tier work.
- Account research and objection handling deserve strong models.
- Anything a rep sends under their name needs a human check.
- Gate CRM writes with permissions and idempotency.
- Measure cost per qualified opportunity, not per message.
How it works, step by step
- Map the sales workflow: enrichment, research, drafting, follow-up, CRM hygiene.
- Route enrichment, deduplication, summarisation and note-taking to cheap tiers.
- Escalate account strategy, competitive positioning and objection handling to strong models.
- Add a quality gate before any human-signed message leaves the system.
- Expose CRM writes as tools with permission checks and idempotency keys.
- Cap sequences per prospect and per day to prevent spray-and-pray spend.
- Track cost per qualified opportunity and reply quality by sequence.
Try it yourself
Open the AI agent cost calculator →
Enrichment, drafting and CRM updates
Most sales-agent work is mechanical: enrich a lead with firmographic data, summarise a thread, log a call note, draft a follow-up. These are bounded, repetitive tasks that cheap tiers handle well, especially with structured output. They also dominate volume, so routing them down is where the budget moves.
Deal-critical judgement is rarer: account strategy, competitive positioning, handling a specific objection, deciding when a prospect is worth escalating. That work changes revenue and deserves a strong model — with a human deciding what actually gets sent.
Route by stage and value
Combine stage and opportunity value into the routing policy:
- Top of funnel: cheap tiers for enrichment, scoring, deduplication and first drafts.
- Qualified opportunities: mid or strong tiers for research and personalised outreach.
- Late stage: strong tiers for objection prep and stakeholder analysis, always human-reviewed.
- Pinned accounts: custom rules so strategic accounts never get downgraded.
Set sequence caps so volume does not silently scale with capability. A cheap model sending ten times more email is not a saving if half the prospects mute the sender.
Quality gates and measurement
Guard the brand: no outreach leaves without a quality check, whether human or a rubric-based automated gate. Validate claims about the prospect against the enriched data, check personalisation fields actually personalised, and block anything that reads as generic filler.
Measure cost per qualified opportunity, reply rate and meeting rate, not cost per email. A cheap sequence with a terrible reply rate costs more in lost pipeline than it saves in tokens. CRM writes should be idempotent and audited so a retry cannot duplicate records or notes. Start on the free plan with plugsky-micro and plugsky-lite for enrichment and drafting, then evaluate strong models for account research during the 14-day full-access trial; plans are on the live pricing page.
Honest comparison
| Sales agent task | Routed sales agent | Strong model for everything | Rules-based sequences |
|---|---|---|---|
| Enrichment and scoring | Cheap tier with schemas | Frontier price per lead | Static fields |
| First-draft outreach | Cheap to mid tier plus quality gate | Native strength | Templates |
| Account research | Strong tier, human reviewed | Native strength | Manual |
| CRM hygiene | Cheap tier plus idempotent tools | Expensive for admin work | Manual or brittle |
| Economics | Cost per qualified opportunity | Cost per message | Cost per send, no automation |
Frequently asked questions
Which sales tasks can run on cheap models?
Enrichment, deduplication, thread summarisation, call notes and first-draft outreach. They are repetitive and structured, and cheap tiers handle them well with JSON output.
Should AI send emails to prospects?
Have a human or an automated quality gate approve outbound messages. Anything sent under a rep's name carries brand risk, and generic AI outreach damages deliverability and reputation.
How do I avoid spammy volume?
Cap sequences per prospect and per day, enforce quality checks, and measure reply and meeting rates. Cost per qualified opportunity exposes sequences that only look cheap.
What belongs in the CRM tools?
Contact updates, activity logging and task creation, exposed with permission checks and idempotency keys so retries cannot duplicate records. Keep bulk destructive operations out of the model's reach.
Which model handles account research?
Use strong tiers for strategy, competitive positioning and stakeholder analysis, and pin strategic accounts with custom rules so they never downgrade.
Is overnight enrichment possible?
The batch endpoint is coming soon. Today, run enrichment in bounded-concurrency workers overnight with a durable queue and per-record status.
How does flat pricing help sales teams?
Self-serve plans are flat monthly with no per-token charges, so testing sequences and prompts does not create a variable bill. See the live pricing page for plan details.
Can I pilot for free?
Yes. plugsky-micro and plugsky-lite are on the free plan with no card, and the 14-day full-access trial covers strong models for account research.