Hand off the whole job
Give it the brief once. It tells you when it'll be done, then does it.
Agents that finish the job.
The harness designed for 400B+ token jobs. Your agents made durable, and ready to handle full projects, not just tasks.
- 95%
- Reliability at a 30-day horizon
- 15x
- Longer autonomous runs
- 80%
- Token usage reduction
- 11.7x
- Fewer hallucinations
Request an invite or ask a friend who's already using Nenya
Used by 4,000+ engineers and researchers at
Same job, same agent. Now more jobs get finished.
Stop working one task at a time. Complete projects with results you can trust.
Claude Code on its own
- draft tuesday's launch post. problem first, then the partner news, 3 bullets, under 200 words. no pricing
- Working…
- Write(launch/post.md)
- Drafted. 310 words, with a pricing line at the end.
- Broke your rule
- too long, and you added pricing. cut it to 200 words and take the price out
- Working…
- Update(launch/post.md)
- Done. 188 words, no pricing. What's next?
- Waiting on you
- You work one task at a time
Claude Code with Nenya
- Launch post
- Press list
- Launch post done
- Socials, Tue to Fri
- Press list done
- Embargo emails
- Nenya: Your last launch sent embargoed posts a day early. Sending Monday 9:00.
- Claude Code wants to email the embargoed post to 30 reporters.
- You allow it with one click
- Partner brief
- You said: 'keep pricing out of anything public'. Post 1 lists $49 a seat.
- Claude Code put pricing in post 1. Taken out, and the posts are back on track.
- A link in the press email pointed to an old page. Fixed and checked before Monday's send.
- Embargo emails done
- Changelog
- Partner brief done
- Launch-day checklist
- Socials done
- Changelog done
- Help docs
- Launch-day checklist done
- Help docs done
- Full projects handled for you
A harness around the agents you use.
Nenya is a harness that works with the models and agent infrastructure you're already using.
Plugs into your stack
Nenya fits in your setup without interrupting your system or your work.
Works with Claude Code, Codex, Cursor, VS Code, and many more.
Your agent
Claude Code, Codex or Cursor
Nenya
- Correct
Catches drift before it turns into a mistake, and steers the agent back on task.
- Remember
Keeps the rules and commitments that matter, so constraints survive compaction cycles.
- Repair
Updates when key details change, and keeps what's still relevant so there's no manual maintenance.
- Non-bypassable rules
Actions with real-world effects pass a boundary your agent can't remove or rewrite.
Your model
On your own keys
Tools and the outside world
Email, CRM, repos, APIs
You keep
- Your agent: Claude Code, Codex or Cursor
- Your model and your own keys
- Your tools, prompts and workflow
- The final say on anything risky
Nenya takes on
- Keeping your rules in force all week
- Catching slip-ups and drift
- Carrying the job across sessions and restarts
- Telling you where the work really stands
Works with any model and any agent stack.
What your agent can do with Nenya
Goodbye context rot
A rule you set on Monday still holds on Friday, however full the context gets.
Catches its own slip-ups
Skipped steps get caught and done, not reported as finished.
Tells you the truth
Gives you honest reports and doesn't invent fake solutions.
Asks before it acts
Anything risky or unclear waits for your call.
Stays inside your limits
Budgets and boundaries hold, even when a pricier option looks better.
Picks up where it left off
A weekend or a crash doesn't lose the thread.
Runs for days on its own
109.6 hours before a person had to step in, against 7.4 without Nenya.
Benchmarks
1.5x the top official score
Remote Labor Index, share of projects accepted
Nenya24.2%
Top official score, July 202615.8%
95.2% still correct on day 30
Task fidelity at day 30, 6,000 runs on three models
Nenya95.2%
Leading agent framework with memory and checkpointing63.9%
Best-practice prompting47.6%
Unaided model41.8%
14.8x longer before anyone steps in
Median hours of uninterrupted operation, 42-day study
Nenya109.6 h
Frontier agent with retry and retrieval7.4 h
All figures are reported in our September 2026 paper, Dawn of Durable Intelligence.
Install
Once you install, Nenya learns your system and handles setup for you.
Request an invite or ask a friend who's already using Nenya
What you can hand off
GTM
Outbound and follow-ups
Research the accounts and chase every reply for weeks.
- Where agents break
- Forgets do-not-contact rules a few days in.
- With Nenya
- Your rules hold for the whole campaign.
Runs forWeeks
Fundraising
Raising your next fund
LP outreach, meetings and diligence requests across the whole raise.
- Where agents break
- Forgets what each LP asked for and what you promised to send.
- With Nenya
- Every LP request and promise stays tracked until it's met.
Runs forWeeks
Research
Long research projects
Research that runs for weeks and builds on itself.
- Where agents break
- Early findings drop out as the context fills.
- With Nenya
- Every finding carries forward, so each week builds on the last.
Runs forWeeks
Operations
Keeping operations in sync
Vendors, budgets, schedules and handoffs that all depend on each other.
- Where agents break
- Changes one thing and misses what it breaks.
- With Nenya
- Knows how every piece connects, so one change carries through to everything it touches.
Runs forWeeks
Regulated work
Compliance and review
Legal review, diligence, coverage checks, grant preparation and government compliance.
- Where agents break
- Hard rules slip on long runs.
- With Nenya
- In our 30-day study across these five workloads, 97.6% of runs broke no hard rules, against 18.4%.
Runs for30 days
Engineering
Long software projects
Migrations and refactors that run across many sessions.
- Where agents break
- Loses the plan between sessions.
- With Nenya
- Picks up where it left off, with the plan intact.
Runs forWeeks
Agents that finish the job.
Hand off the whole job. Get it back completed.
Request an invite or ask a friend who's already using Nenya