The control plane for AI agents.
HKT records, monitors, and constrains agent workflows. Blackbox tracing, OS-level safety, and cost guards — without slowing your agent down.
Agents break. Dashboards show you the wreckage, not the cause.
Constraint Decay
Long sessions lose context. Important instructions get compacted away. Your agent forgets what it was told and repeats itself.
Safety Evaporation
Prompt-layer safety controls aren't real boundaries. Agents find alternative paths through other tools.
Memory Contamination
Persistent memory acts like an ungoverned database. Preferences bleed across users. Sessions leak between projects.
Infinite Loops
Expired tokens, failed tool calls, and context overflow trigger endless retry loops that burn tokens and crash sessions.
Lost in the Codebase
Agents follow imports indefinitely, burning timeout on exploration instead of converging on a plan.
Blind Debugging
Dashboards show that a run failed. They can't tell you which step, which agent, or what missing state caused it.
Three steps to control your agents.
Connect
Drop HKT into your existing workflow. One env var. Works with Claude Code, Cursor, Hermes, and any OpenAI-compatible agent.
Observe
Every agent action gets recorded, classified, and scored. See what's happening in real time on a live dashboard.
Control
Set budgets, enforce safety rules, catch loops before they burn tokens. Intervene automatically, not manually.
Three tools, one runtime.
Blackbox records, Shield enforces, Cost Guard kills loops. Run them together or pick one.
Agent Blackbox
See everything your agent does
Record every agent run: prompts, tool calls, file changes, commands, errors, decisions. Replay-grade traces, not just chat history.
- ✓Full prompt & response logging
- ✓Tool call chain visualization
- ✓File diff tracking with rollback
- ✓Timeline replay of entire sessions
Agent Shield
Real boundaries, not prompt suggestions
Real safety checks before agents edit files, run commands, or call APIs. Enforced at the OS and network level.
- ✓OS-level file write guard
- ✓Network request filtering
- ✓Command execution sandbox
- ✓Per-tool permission policies
Cost Guard
Stop burning money on broken loops
Detect loops, wasted retries, and runaway token usage before they become expensive. Cost per successful task, not just total spend.
- ✓Real-time loop detection
- ✓Token budget per task
- ✓Cost-per-success metrics
- ✓Automatic circuit breaker
Start free. Lock in founder pricing.
Early adopters keep permanent pricing — even when we ship paid tiers.
For individuals exploring agent observability.
- 3 agents
- 7-day trace retention
- Cost Guard basic
- Community Discord
For indie hackers and builders shipping daily.
- Unlimited agents
- 90-day trace retention
- Cost Guard + Shield
- Priority email support
- Self-host or cloud
For teams running agents in production.
- SSO + audit log
- SLA + dedicated support
- On-prem deployment
- Custom integrations
- Everything in Pro
Questions, answered.
What agents does HKT work with?
Any agent that makes API calls, runs shell commands, or edits files. First-class integrations for Claude Code, Cursor, Hermes Agent, OpenAI Codex. OpenAI-compatible proxy for everything else.
Is this another monitoring dashboard?
No. Dashboards show what happened. HKT prevents things from happening. Agent Shield enforces rules at the OS level, Cost Guard kills runaway loops in real time, Blackbox gives you replay-grade traces.
How much does it cost?
Early access. Pricing hasn't been finalized, but early adopters lock in permanent founder pricing. Join the waitlist.
Does it slow down my agent?
Under 2ms overhead per tool call. Async logging and non-blocking safety checks. Your agent won't notice.
Can I self-host it?
Yes. Runs locally or in your own cloud. No data leaves your machine unless you opt into cloud features.
Stop your agents from burning the runway.
Join the waitlist. Lock in founder pricing. Ship agents you actually trust.
No spam. We email only when there's something to ship.