AI agents, in production.
Now booking Q3 engagements

Anyone can demo an LLM. We ship the system that survives production.

[About]

Metaheuristic is an applied AI agency. We take teams from "we tried ChatGPT" to agents that take real actions across your CRM, tickets, docs, and APIs - with permission-aware RAG, evals, guardrails, and the LLMOps to keep them reliable, secure, and on budget.

planacttoolapprove
Agentic workflows
Plan · act · tool calls · human approvals
contracts/2026-q2.pdf allow ✓
finance/payroll.xlsx deny ✕
wiki/runbooks allow ✓
hr/reviews deny ✕
Cites
what you can see.
ACL-filtered retrieval
Permission-aware RAG
Citations · ACLs · freshness · evals
traces evals prompts versions routing cost latency drift alerts
Run it
like production.
LLMOps / AgentOps
Monitoring · evals · versions · routing
LLM01 · prompt injection guarded
LLM02 · insecure output guarded
LLM04 · data poisoning guarded
LLM05 · supply chain review
OWASP LLM Top 10
red-teamed
LLM security review
Injection · output handling · supply chain
offline online CI gate
Evals & quality gates
Golden sets · graders · CI gates
cost · this month
$4,120 → $1,180
−71% · caching + routing
cache route
Burn
less on tokens.
AI cost optimization
Caching · routing · context budgets
route(task) {
  if (simple)   → haiku
  if (reason)   → sonnet
  if (critical) → opus
  if (cached)   → return hit
}
Right model.
Right price.
Model routing
Per task · per cost · with fallback
trace · latency by stepp95
retrieve · reason · act
span-level traces
tokens · latency
· cost
Observability & tracing
Spans · tokens · cost · per step
Pending Approved Blocked
act · refund needs approval
issue $480 refund · order #2841
tool: stripe.refunds.create
act · email approved
send follow-up to lead
tool: crm.send_email
act · delete blocked
drop records · out of policy
guardrail: destructive
Human-in-the-loop
Approvals · safety limits · logging
Aug 2024 · entered into force
Feb 2025 · prohibited practices
Aug 2025 · GPAI obligations
Aug 2026 · broad applicability
Compliant
by design.
EU AI Act readiness
Risk class · docs · governance
agentCRMAPIsERPdocs
Across
your stack.
Tool & system integrations
CRM · ERP · tickets · docs · APIs
requestretrievereasontool_callapproveacteddeferredblocked
Audit logs & lifecycle
Every step. Every action. Replayable.
[Process]

Audit in. System out.

01 · Audit
Map & prioritize

We map workflows, assess data readiness, and rank opportunities by ROI and risk.

02 · Blueprint
Architecture + prototype

A clear technical plan and a working prototype - before you commit to a production build.

03 · Build
RAG & agents

Permission-aware RAG or agentic workflows with guardrails, memory, and integrations.

04 · Evaluate
Evals + security

Golden-set evals, retrieval quality, and an OWASP LLM red-team before anything ships.

05 · Deploy
Approvals + logging

Ship with human approvals, safety limits, and full audit logs. Code ownership stays with you.

06 · Operate
LLMOps retainer

Monitoring, evals, prompt/version management, model routing, and cost control.

Agentic workflows· Permission-aware RAG· LLMOps / AgentOps· EU AI Act ready· OWASP LLM Top 10· Cost optimization· Evals + guardrails· Agentic workflows· Permission-aware RAG· LLMOps / AgentOps· EU AI Act ready· OWASP LLM Top 10·
[Engagements]

Start small. Scale to production.

AI Workflow Audit
from $9,500
1–2 weeks · diagnostic
  • · Map workflows + data readiness
  • · High-ROI opportunity ranking
  • · Prioritized implementation plan
Book an audit
AI System Blueprint Sprint
from $22,500
2–4 weeks · architecture
  • · Target architecture + risks
  • · Working prototype
  • · Success metrics + roadmap
Scope a sprint
Most requested
Production RAG / Copilot
from $45,000
secure knowledge system
  • · Permissions + citations
  • · Retrieval evals + freshness
  • · Monitoring + audit logs
Talk to us
Agentic Workflow MVP
from $75,000
agents that take action
  • · Controlled actions across tools
  • · Human approvals + safety limits
  • · Logging + observability
Talk to us
LLMOps / AgentOps Retainer
from $7,500 / mo
recurring · on-call
  • · Monitoring + evals + routing
  • · Prompt / version management
  • · Cost control + governance
Start a retainer
Not sure where to start?
A 30-minute call and a fixed-scope audit usually answer it.

Clear scope, milestones, code ownership with you, and measurable success metrics on every engagement.

Book a call

Fixed scope and milestones. You own the code. Volume and enterprise terms (compliance, multi-agent, production hardening) scoped on request.

[Landscape]

Most shops ship a demo. We ship a system you can run.

At a glanceMetaheuristicIn-house hireGeneric dev shopNo-code (Zapier/Make)Off-the-shelf copilotDIY ChatGPT
What you getProduction systemCapacity over timeA build, maybeBrittle automationsGeneric chatA prototype
Time to production2–10 weeksMonths to hireMonthsDays, then stuckInstantNever quite
Guardrails & approvalsBuilt inDepends--Vendor-set-
Evals & securityEvals + OWASP LLMDependsRarely-Black box-
Cost controlRouting + cachingAd hocAd hocPer-task fees$ / seat / moUnmanaged
OwnershipYour code + IPYoursNegotiatedLocked-inVendorYours
Ongoing opsLLMOps retainerOn the teamHand-offYou maintainVendor SLAYou maintain

A demo proves the model can do it once. We build the guardrails, evals, security, and ops that make it hold up every time - and we stay on to run it.

[How it runs]

An agent that asks before it acts.

support-agent · trace
user› "Customer #2841 was double-charged. Fix it."

retrieve› permission-aware RAG · 3 sources
  - orders/2841.json        allow ✓
  - policy/refunds.md       allow ✓
  - finance/ledger.xlsx     deny ✕ (out of scope)

reason› duplicate charge confirmed · within policy
tool› stripe.refunds.create({ amount: 480_00 })

⏸ guardrail: refund > $200 needs human approval
✓ approved by jordan@acme · 41s

› refund issued · customer notified · logged
$ 
[Audit log]
{ "action": "refund",
  "amount": 48000,
  "approved_by": "jordan",
  "sources": 2,
  "status": "completed" }
signed · immutable · replayable
[Eval scorecard]
groundedness0.98 tool accuracy0.96 refusalsok
gated in CI before deploy
Book an AI Workflow Audit Talk to us applied AI agency · metaheuristic.co