Hebbrix
Hebbrix is persistent agent memory that ranks recall by reported outcomes, not text similarity.
Outcome-weighted recall is the right fix for agents that keep walking back into the same wall, and Hebbrix is built around that idea rather than bolted onto a vector store. The one-line base-URL swap and a 1,000-credit/month free tier make the trial cheap, and API v1.1.3 (October 2026) shows the ranking machinery still moving with opt-in selection methods and explicit index waits. The caution is that the vendor withdrew its public LoCoMo claim and its own changelog calls recent releases engineering corrections rather than proof of improved performance. Measure recall on your own traffic before scaling. If your agent can report outcomes, Hebbrix is worth a serious trial; if it can't, Mem0 or
Verified 3d ago · liveness 78/100 · cite: rightaichoice.com/tools/hebbrix
- Developers running production agents with a real feedback loop to report outcomes into
- Support and voice agents where a repeated mistake shows up in a customer-facing metric
- Teams on LangChain, LangGraph or CrewAI wanting drop-in persistent memory
- Multi-tenant platforms needing audit-ready, exportable, model-agnostic memory per tenant
- Single-session chatbots with no cross-session memory requirement at all
- Buyers whose agent cannot report outcomes back — outcome-weighted ranking has little to rank
- High-frequency, low-value apps where credit-based billing costs more than the memory is worth
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Hebbrix if your agent has no mechanism to report delayed outcomes back — outcome-weighted ranking has almost nothing to rank without feedback, and a plain vector store or Mem0 covers the same ground for less.
On Free and Starter, memory writes pause when monthly credits run out while reads continue, so a busy agent can silently stop learning mid-cycle.
The credit ladder maps cleanly to stage: Free ($0/mo, 1,000 credits) for a hobby project, Starter ($19/mo, 25,000) to ship a first product, Pro ($99/mo, 200,000) for a production assistant serving a few thousand daily users, Scale ($399/mo, 1,000,000) for high-traffic multi-tenant platforms, and Enterprise pay-as-you-go above that. Against Mem0 and managed vector stores the entry cost is comparable; the real pricing decision is whether your agent reports outcomes well enough to justify memory
In short
Hebbrix — Hebbrix is persistent agent memory that ranks recall by reported outcomes, not text similarity. Best for Developers running production agents with a real feedback loop to report outcomes into, Support and voice agents where a repeated mistake shows up in a customer-facing metric, Teams on LangChain, LangGraph or CrewAI wanting drop-in persistent memory. Free to start; paid plans from $19/mo.
What's new in Hebbrix
Checked 3 days agoAcross the latest 5 updates: 4 changelog entries and 1 news mention.
API v1.1.3 / customer-round12-v2-20261007-2c27596e2 Fix Missing-source Recall and Exact Indexing Observation
Unsupported personal recall can now abstain before generation instead of guessing, and /memories/process accepts explicit wait_for_index with a bounded index_wait_seconds observation. Published Python 2.6.2, TypeScript 2.5.2 and MCP 0.5.14 are unchanged.
API v1.1.3 / Round10 engineering contracts Fix Source-time, Scope and Response Corrections
Personal collection reads now require an explicit end-user scope, and memory-off chat preserves caller instructions without hidden profile injection while disclosing requested and served model identities.
API v1.1.3 / customer-round11-20261007-a37b02820 Fix Committed Memory Receipts and Explicit Learning Waits
Memory processing acknowledges the exact committed source and indexing event, and a durable pending receipt counts as a successful write rather than a searchable completion. SDKs remain Python 2.6.2 and TypeScript 2.5.2.
API v1.1.3 / customer-round9-20261006-18a5e141b / Python 2.6.2 / TypeScript 2.5.2 Fix Explicit Evidence, Readiness and Native Selection Contracts
Adds compact evidence views, bounded chat-learning waits, owned batch receipt reads, explicit provisional-outcome confirmation and native selection controls. A timed-out wait is not completion.
A memory write is not a retrieval guarantee
Engineering guide on building a durable read-after-write workflow with explicit indexing receipts and safe retries, published alongside companion posts on safe agent memory boundaries and avoiding double-counted outcome learning.
Viability Score
How well maintained and how widely used is Hebbrix? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- Outcome-weighted recall ranks memory by reported success or failure, not text similarity
- Five retrieval layers: adaptive semantic, lexical, knowledge graph, temporal and feedback signals
- One-line OpenAI drop-in: change base_url to api.hebbrix.com/v1
- Chat completions endpoint with automatic memory retrieval and context injection
- Outcome Memory API with decision receipts, delayed outcomes and posterior uncertainty
- Knowledge graph with entity extraction, relationship mapping and fact-replacement tracking
- 3-tier memory with automatic promotion across short, mid and long-term stores
- Memory decay driven by usage, recency, importance, corrections and reported outcomes
- Missing feedback treated as unknown rather than failure
- Collections for organizing memories in multi-tenant applications
- MCP server 0.5.14 with 33 live tools for MCP-compatible agents
- Coding agent MCP integration for Claude, Cline and other MCP clients
- Explicit /memories/process index waits (wait_for_index, bounded index_wait_seconds)
- Opt-in selection methods with experimental outcome-policy-v3 ranking
- Memory-off chat that preserves caller instructions without hidden profile injection
About Hebbrix
Hebbrix is a persistent memory layer for AI agents that ranks what to recall by whether a memory led to a real outcome, rather than by how close it sits to the current query. The company frames the gap plainly: similarity search answers "what looks like my query?" and temporal validity answers "is this fact still true?" — neither asks whether acting on the memory actually worked. Hebbrix asks that third question, and its docs treat detecting failure as the easy half, with keeping what worked being where memory systems quietly break. Integration is a one-line swap: you point an existing OpenAI-compatible client at api.hebbrix.com/v1 and memory is retrieved and injected into context automatically, with no extra tool calls, no migration and no rewrite. Under the hood, recall blends adaptive semantic, lexical, knowledge-graph, temporal and feedback signals, and a 3-tier store promotes memories across short-, mid- and long-term layers. Decay is driven by usage, recency, importance, corrections and reported outcomes, and the docs state that nothing is silently deleted just because no outcome arrived. Outcome reporting is the differentiator: your agent files a delayed success, failure, business metric or correction against a decision receipt, and Hebbrix updates tenant-scoped posterior evidence while treating missing feedback as unknown rather than failure. An MCP server exposes 33 live tools to MCP-compatible agents, with a stated coding-agent integration for Claude and Cline. Stored memory is model-agnostic and portable — export every memory and the knowledge graph on any plan. Pricing is credit-based and monthly: Free at $0/mo with 1,000 credits, Starter at $19/mo with 25,000, Pro at $99/mo with 200,000, Scale at $399/mo with 1,000,000, plus a pay-as-you-go Enterprise tier with SSO and procurement review. Choose it over Mem0 or a plain vector store when you genuinely have a feedback loop to feed it.
Behind the Verdict
Hebbrix's core argument is easy to state and hard to argue with: retrieval systems optimize for resemblance when what an agent needs is the memory that previously produced a good result. The architecture reflects that. Five retrieval layers (adaptive semantic, lexical, knowledge graph, temporal, feedback) rank evidence, and the Outcome Memory API lets an agent file a delayed success, failure, business metric or correction against a decision receipt so tenant-scoped posterior evidence updates. Missing feedback is treated as unknown rather than failure, which matters — a support agent that simply forgot to report back shouldn't be punished as if it failed. The drop-in story is genuinely convenient. You change one line in an OpenAI client (base_url to api.hebbrix.com/v1) and memory retrieval plus context injection happens inside the chat-completions call you already make. No extra tool calls, no rewrite. On top of that, an MCP server with 33 live tools covers MCP-compatible agents, including a stated coding-agent path for Claude and Cline. The 3-tier memory store promotes across short, mid and long-term layers, and decay is driven by usage, recency, importance, corrections and reported outcomes. Export of every memory and the knowledge graph works on any plan, so you are not locked in by data gravity. Where you should be careful is evidence quality. The company withdrew its public LoCoMo benchmark claim, and recent changelog entries are unusually candid that they are engineering corrections rather than new performance evaluations. Several entries explicitly say benchmark and evaluation material remains private, and that ranking scores are not correctness certificates. That is honest engineering writing — and also a signal that you should build a small eval set and measure recall on your own traffic rather than trust a headline number. Billing is the other friction point. Credits are normalized usage units, not 1:1 API calls, so you must estimate quota from documented credit weights and the dashboard. On Free ($0/mo, 1,000 credits) and Starter ($19/mo, 25,000 credits) memory writes pause when credits run out while reads continue; uninterrupted overage billing only exists on Pro ($99/mo) and Scale ($399/mo). The pricing page's own worked examples — roughly 300 chat turns for 1k credits, a few thousand daily users for 200k — are the ones to sanity-check against your load. Enterprise is pay-as-you-go with SSO and a security questionnaire, not a published list price. Fit-wise: strong for production agents with a real feedback loop, support and voice agents where a repeated mistake shows up in a customer metric, and teams on LangChain, LangGraph or CrewAI; weak for single-session chatbots and any agent that cannot report outcomes back.
Researching Hebbrix? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas Hebbrix actually fits — and what changes day-one when you adopt it.
Swap the OpenAI base_url to api.hebbrix.com/v1, create a collection for the tenant, and have the agent file a success or failure receipt after each resolved ticket.
Outcome: The agent stops re-suggesting fixes that already failed for that customer, and recall skews toward resolutions with reported outcomes.
Use Collections plus an explicit end-user scope for personal reads, keep memory-off chat paths for requests that must not inject a profile, and wire export into the tenant offboarding flow.
Outcome: Each tenant's memory stays separated and auditable, and offboarding satisfies export and deletion requirements without a data migration.
Connect the Hebbrix MCP server (33 live tools) to Claude or Cline, then report outcomes when a fix holds or a build breaks.
Outcome: The coding agent keeps the debugging detail that actually unblocked the build instead of re-trying the same dead-end patch.
Use Cases
- Store user preferences across sessions so your assistant personalizes without re-prompting.
- Let support agents recall past issues and the resolutions that actually worked.
- Build an internal knowledge base that learns from reported outcomes instead of accumulating text.
- Give voice assistants context that survives across separate calls.
- Power sales agents that track prospect interactions and adapt pitches over time.
- Give a coding agent MCP-connected memory so it stops revisiting failed fixes.
- Keep tenant-scoped memory separate in a multi-tenant platform with per-tenant export.
Models Under the Hood
as of 2026-10-18
Limitations
- Credits measure normalized memory work rather than 1:1 API calls, so usage against a plan's quota must be estimated from documented credit weights and the dashboard; the pricing page's own examples (roughly 300 chat turns for 1k credits, a few thousand daily users for 200k) are the yardstick.
- On Free ($0/mo) and Starter ($19/mo), memory writes pause once monthly credits are exhausted while reads continue; overage billing that avoids interruption is only available on Pro ($99/mo) and Scale ($399/mo), adding variable cost.
- Hebbrix's own changelog is explicit that recent releases are engineering corrections, not proof of improved performance, and that benchmark, evaluation and raw validation material remains private.
- Missing feedback is treated as unknown rather than failure, and API timeouts or pending receipts do not indicate successful completion — a durable pending receipt is a successful write, not a searchable completion.
- Graph reads can appear on lower tiers, but manual graph mutation and inference have additional gates described in the graph guide.
as of 2026-10-16
Verification history
We have re-verified Hebbrix 9 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-checked, vendor evidence unchanged
Showing the 6 most recent of 9 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Plans compared
For each published Hebbrix tier: who it actually fits, and what it adds vs. the previous tier. Cross-reference the cost calculator above for projected annual outlay.
Free
$0/mo
Ideal for
Developers and hobby projects testing whether cross-session agent memory is worth wiring up, with no credit card required.
What this tier adds
Starting tier: 1,000 credits/month, adaptive hybrid search, profile extraction and community support; memory writes pause when quota runs out while reads continue.
Starter
$19/mo
Ideal for
Small teams shipping a first product that needs memory to survive between sessions.
What this tier adds
Adds the knowledge graph and bring-your-own model key on top of Free, at 25,000 credits/month.
Pro
$99/mo
Ideal for
Production assistants serving a few thousand daily users where writing pauses are unacceptable.
What this tier adds
Adds the Outcome Memory API, advanced analytics and overage billing so usage is never interrupted, at 200,000 credits/month.
Scale
$399/mo
Ideal for
High-traffic agents and multi-tenant platforms pushing past a million credits of monthly memory work.
What this tier adds
Raises the quota to 1,000,000 credits/month and lifts rate limits, keeping overage billing from Pro.
Enterprise
Pay as you go
Ideal for
Organizations with compliance, procurement and single-sign-on requirements that need custom contract terms.
What this tier adds
Pay-as-you-go pricing with bring-your-own keys for any model, a security questionnaire and procurement review, solutions engineering and SSO.
Where the pricing makes sense
The company stage and team size where Hebbrix's pricing actually pencils out — and where peers do it cheaper.
The credit ladder maps cleanly to stage: Free ($0/mo, 1,000 credits) for a hobby project, Starter ($19/mo, 25,000) to ship a first product, Pro ($99/mo, 200,000) for a production assistant serving a few thousand daily users, Scale ($399/mo, 1,000,000) for high-traffic multi-tenant platforms, and Enterprise pay-as-you-go above that. Against Mem0 and managed vector stores the entry cost is comparable; the real pricing decision is whether your agent reports outcomes well enough to justify memory
Setup time & first value
How long it actually takes to get something useful out of Hebbrix — broken out by persona, not the marketing-page minute.
One line of code if you already call an OpenAI-compatible chat endpoint — change base_url to api.hebbrix.com/v1 and memory retrieval happens inside the same call, so first recall is typically same-day. SDK users (Python 2.6.2, TypeScript 2.5.2) get collection creation and memory writes within the first hour. Wiring outcome reporting and a small eval set to judge recall on your own traffic is the
Switching to or from Hebbrix
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From Mem0: point your OpenAI client at Hebbrix's base_url and recreate memories via the create and search endpoints or collections API.
- →From a plain vector store: move text into Collections via client.memories.create, letting the knowledge graph and outcome layers layer on top.
- →From Anthropic-based agents: the docs state Hebbrix works with OpenAI, Anthropic and any LLM — change the base URL and keep your stack.
- →From a custom SQL memory table: export rows and load them as memories with collection_id scoping per tenant.
- ↗To Mem0: export every memory and the knowledge graph on any plan, then load the text and relationships into Mem0.
- ↗To a plain vector store: use the any-plan export to pull memory text and rebuild embeddings in your own index.
- ↗To another LLM provider: no migration needed — stored memory is model-agnostic, so switch the underlying model and keep the memory.
- ↗To an in-house store: exported memory plus knowledge graph gives you the raw entities and relationships to self-host.
Integrations
Resources & Guides
- Documentationhebbrix.com
Docs · Hebbrix
Full product docs from hebbrix.com
- Quickstarthebbrix.com
Quickstart · Hebbrix
Get up and running fast from hebbrix.com
- Documentationhebbrix.com
Coding Agent · Hebbrix
Full product docs from hebbrix.com
- Documentationhebbrix.com
Multi Tenancy · Hebbrix
Full product docs from hebbrix.com
- Resourcehebbrix.com
Changelog · Hebbrix
Helpful link from hebbrix.com
Tutorials & Learning
YouTube returned 6 videos for “Hebbrix”, and we withheld 6: 6 could not be judged, because “Hebbrix” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about Hebbrix.
Official links
Featured Head-to-Head Comparisons
Hebbrix vs Spider Cloud
Spider Cloud and Hebbrix solve different problems. If your AI agent needs to fetch fresh web data for RAG or scraping, Spider Cloud’s Rust engine and Browser AI commands are unmatched. If you need persistent, self-improving memory for agents (customer support, voice, etc.), Hebbrix’s outcome-weighted recall and knowledge graph are a game-changer. Choose based on your primary need: external data or internal memory.
Hebbrix vs Presto Voice
Presto Voice and Hebbrix serve entirely different markets: Presto is a turnkey drive-thru voice AI solution for QSR chains, while Hebbrix is a developer-focused memory layer for building any AI agent. Choose Presto if you run a multi-location fast-food chain and want proven revenue lift. Choose Hebbrix if you're an AI developer needing persistent, outcome-weighted memory that integrates via one line of code.
Hebbrix vs Temporal Ai
Temporal is the clear choice if you need durable, fault-tolerant orchestration for complex multi-step workflows, especially AI agents that must survive crashes. Hebbrix excels at adding smart, persistent memory to existing agents, but if you already need orchestration, Temporal's memory is built-in. For a pure memory layer on top of a separate orchestrator, Hebbrix is drop-in easy. For everything-in-one, go Temporal.
Popular in Agent Memory & Runtimes
Frequently Asked Questions
Categories
Best-of guides
Used Hebbrix? Help shape our editorial sentiment research.