LEGION CODE × ACTIVELOOP

Memory that
sticks.

honeycomb runs underneath your coding assistant as a local daemon. It captures every turn, distills it into source-backed memory, and recalls the right context before the next turn, across every tool.

$ curl -fsSL https://get.theapiary.sh | sh
macOS · Linux · Windows read the docs →
honeycomb · recall ● daemon :3850
recall ""
4 hits · 0.94 top
deploy-staging.md ✓ verified
push to staging via the deploy script
ops/deploy-staging.md:12 personal 0.94
session 8f2a
staging uses the eu-west cluster
session 8f2a · 2026-06-19 personal 0.88
release.yaml
smoke test gates the staging deploy
ci/release.yaml:40 personal 0.81
0007-auth.md ✓ verified
supabase auth + rbac
adr/0007-auth.md:3 personal 0.92
refresh-token.ts
refresh handled server-side
auth/refresh-token.ts:42 personal 0.86
session 4c19 dreaming…
consolidated from three sessions
session 4c19 · 2026-06-11 personal 0.79
limits.ts ✓ verified
100 req/min per token
api/limits.ts:18 personal 0.90
everyday-use.md
limits apply per workspace
docs/guides/everyday-use.md personal 0.84
session 1ad7
raised limit for batch jobs
session 1ad7 · 2026-06-02 personal 0.77
THE PROBLEM

Your AI agents forget everything.

Close the window and the context is gone. Open a different tool tomorrow and it has never heard of your project. So you re-explain your conventions and re-discover fixes that already worked.

Context dies with the window
Every tool starts from zero
The team can’t see what you learned
How it works

Capture. Distill. Recall. Compound.

One simple loop runs quietly in the background. The more you use it, the sharper it gets, instead of noisier.

01

Capture

Notice what happens as you work: prompts, actions, results. Cheap, instant, never in the way.

02

Distill

Turn raw transcripts into clean notes: a one-line key, a short summary, the full original on demand.

03

Recall

Hand the right notes back at the start of every session, matched by your words and by meaning.

04

Compound

Tidy itself over time: merge duplicates, prune junk, supersede stale facts. Sharper, not noisier.

27.5% avg.avg.
reduction in token spend
1.4x avg.avg.
faster llm responses
6+
harnesses supported
Source-backed memory

Receipts, or it doesn't count.

Every recalled memory shows its source (a file path or a session id) and a relevance score. Verified means a human-checked, source-backed fact. No black box, no guessing.

Provenance on every claim
Hybrid lexical + 768-dim semantic recall
Three tiers: key, summary, raw. Zoom only as far as you need
deploy/prod-022 ✓ verified
We deploy from the release branch, never from main.
infra/deploy.md personal 0.94
skill/pr-review
Reusable PR-review checklist, mined from 9 sessions.
session 4b81 personal 0.92
graph/build-order dreaming…
Consolidating 3 overlapping notes into one current fact…
dreaming personal 0.86
The Pollinating loop

While you sleep,
it gets smarter.

Every so often Honeycomb runs a self-tidying pass: merging duplicates, pruning junk, replacing stale facts with their current version, keeping a full history so nothing is truly lost. Most note piles get messier as they grow. Honeycomb does the opposite.

merged5 duplicates → 1
supersededprod-018 ⇒ prod-022
skillifieddeploy/rollback
One memory, every tool

Works underneath six
coding assistants at once.

A note written while using one assistant is recalled by another. Switch tools mid-project and the context comes with you.

Claude Code ● connected
Cursor ● connected
Codex ● connected
Hermes ● connected
pi ● connected
OpenClaw ● connected
● org · acme
● workspace · backend
● project · api-gateway
deploy/prod-022shared
skill/pr-reviewteam
For teams

One brain for the whole team.

A teammate solves a tricky migration on Monday; the reusable lesson reaches everyone's assistants by their next session, no file passed around. Memory stays shared and separated, scoped cleanly across org, workspace, and project.

Skills that spread automatically
Publish a skill and it picks up in every teammate's assistants at the start of their next session, carrying who wrote it.
Private by default, widened on purpose
When in doubt, Honeycomb shows less, not more. You promote something to the whole team as a deliberate, recorded action.
The dashboard

See everything it remembers.

A simple local web page: your memories, connected tools, shared skills, a map of your codebase, and the health of everything. No database knowledge required.

honeycomb · dashboard ● all systems healthy
Memories
1,284
▲ 12 this week
Sessions
312
Recall score
0.94
Skills
47
▲ 8 this week
Memories captured · 14d
Live log
captured · cursor · 2 turns
recalled · 4 hits · 0.94
pollinated · merged 5 → 1
skillified · pr-review
Your data stays yours

Private by default. Your store, your rules.

Memories live in your own Deep Lake store, separated cleanly by team and project, versioned and inspectable. Secrets are kept apart and never shown to an assistant.

Your store, powered by Deep Lake

The local daemon is the only thing that connects to it. Host it in your own cloud if you want.

Loopback only

On a single machine the daemon binds 127.0.0.1:3850, and listens only to your own computer.

Versioned & inspectable

Full history on every memory. Sharing is opt-in by design; nothing bleeds across lanes.

Give your agents a memory.

One command. The daemon binds loopback only, captures every turn, and recalls source-backed memory before the next, across every harness, for you and your team.

$ curl -fsSL https://get.theapiary.sh | sh
Windows (PowerShell): irm https://get.theapiary.sh/install.ps1 | iex