Connected knowledge powers today's agent.
Memco governs what tomorrow's agent inherits.
Onyx connects workplace sources to enterprise search, AI assistants, agents and actions. Memco is built around the next layer: turning corrections and outcomes from real agent work into scoped lessons a later person or agent can trust.
Different person · Different session · Different tool · Reused lesson
The stakes
Your team already paid to learn the lesson. The next agent should not pay again.
Coding agents discover repository quirks, rejected approaches and working fixes every day. The useful part is rarely the whole transcript. It is the lesson that should change a later run, attached to the source, scoped to the right team and kept current when the code changes.
Fewer repeated mistakes
Later agents receive the known constraint before they repeat a failed approach.
Less senior re-explanation
Corrections stop dying in review comments, chat threads and one engineer's memory.
Cross-tool continuity
The lesson survives when the team moves between Claude Code, Cursor, Codex, Copilot or internal agents, subject to actual integration support.
Customer-owned learning
The organization keeps the lesson, source and scope rather than renting the same rediscovery from each model or tool.
01Honest overlap
Where the products genuinely overlap.
Why this is a real comparison.
- 01Both products make company-specific knowledge available to AI and care about permissions, model choice and enterprise deployment.
- 02Onyx is not merely a search box: it has custom agents and actions, open-source self-hosting, broad connectors, an MCP server surface and a beta coding agent.
- 03The clean comparison is read/action path versus governed learning path — not retrieval versus intelligence.
02Where each product starts
Different centers of gravity.
Onyx centers the permission-aware AI, search and action surface over connected company knowledge. Memco centers the governed conversion of new work into reusable learning.
Onyx
Find, synthesize and act on information across company sources through enterprise search and AI assistants.
- Primary inputs: documents, messages, tickets, code, CRM data and other connected workplace sources.
- Outcome: a person or agent can find grounded evidence and act through connected tools.
Memco
Convert real agent work into reviewed, scoped and reusable team lessons.
- Primary inputs: agent attempts, human corrections, failed paths, tests, review decisions, accepted fixes and downstream outcomes, according to released integration behavior.
- Outcome: a correction or outcome from new work becomes a scoped lesson that changes a later run.
03The visual argument
What can we find now, or what should the organization carry forward next?
Both loops are real operating models. The difference is direction: Onyx reads and acts over what is already connected; Memco writes what just happened into a governed lesson and closes the loop when a later run proves it.
The Memco track carries the states the organization keeps: source · scope · review · later outcome
04The same week, replayed
One payment-retry bug. Two product jobs.
Illustrative engineering workflow, not customer evidence — the same payment-retry scenario runs on every comparison page, so what differs is the product job, not the anecdote.
- 01Agent A · Cursorpayment-retry taskcalls a helper that was retired last quarter
- 02Failed testci · redthe dead end costs the run its first hour
- 03Senior correctionreview + passing PRthe durable fact: which path works now, and why
- 04Scoped reviewed lessonteam scope · provenance keptsource, reviewer, scope and outcome attached
- 05Agent B · Claude Codenext week · different personstarts ahead of the dead end
Onyx can retrieve the relevant repository documents, PR discussion and connected evidence, then let an agent act through configured tools.
Memco focuses on promoting the correction and passing outcome into a scoped lesson, then proving that a later agent used it before repeating the failure.
05Practical comparison
Dimension by dimension, in words.
Rows, not scoreboards: each dimension states what each product actually does, so the difference survives scrutiny.
06Choose honestly
Two good answers to two different questions.
Choose Onyx if
- Company knowledge is scattered across connected workplace sources.
- Permission-aware enterprise search and grounded answers are the main problem.
- Teams want custom assistants and actions over that indexed knowledge.
- Open-source / self-hosted workplace AI is the desired product surface.
- The main job is to find, synthesize and act on information across company sources through enterprise search and AI assistants.
Choose Memco if
- A lesson from one person's agent should improve a later task for another person or tool.
- Corrections, failed paths and accepted outcomes keep getting rediscovered.
- Shared learning needs source, scope, review and a path to correction or retirement.
- The organization wants its learning to survive model, IDE and agent-platform changes.
- The job is to make real engineering learning travel safely across people and tools — run the reuse test.
07Proof, not vibes
The reuse test.
Do not stop at "the agent remembered." Onyx's proof question is: did the assistant retrieve grounded evidence, answer accurately and complete the configured action? Memco's is: did a source-backed lesson from one run improve a later task for another person, agent, session or tool without violating scope?
A signup is not activation. Activation is a lesson contributed from one task and correctly reused on a later task.
08Enterprise confidence
Governed by construction.
Shared memory becomes infrastructure. Treat it like infrastructure. Private company memory stays inside your organisation — there is no public pool for it to reach.
09Questions
What evaluating teams ask.
Source ledger
Statements about Onyx are drawn from its official documentation and site, reviewed on 2026-07-30. Products change — if anything here is out of date, tell us and we will correct it.
- 01Onyx — open-source enterprise search and AI assistant positioning
- 02Onyx — custom agents and actions
- 03Onyx — current integrations
- 04Onyx — current pricing and deployment
- 05Onyx docs — connector model
- 06Onyx docs — agent features
- 07Onyx docs — action features
- 08Onyx docs — beta coding-agent surface
- 09Onyx docs — MCP server surface
- 10Onyx docs — audit logs
- 11Onyx docs — access controls
- 12Onyx docs — cloud compliance
- 13Onyx — current license split
- 14Memco — product positioning and shared-memory focus
- 15Memco — team pricing and memory controls
- 16Memco — product and integration documentation
- 17Memco — making sense of agentic memory
- 18Memco — active memory: reinforcement learning for agents
- 19Memco — the art of forgetting
Test whether one lesson can improve the next run.
the loop
benchmarks · product · research
A short dispatch on shared memory for AI agents — the numbers behind the product, what we're shipping, and the research we're reading. No filler.