See the whole team at a glance
Every engineer’s sessions on one live timeline, colored by judged outcome — who’s shipping, what’s stuck, and where your agents spend their time. Hover any block; click to inspect.
Every session becomes a trace
The moment an agent runs, Octarin captures it — files, commands, cost, and an AI summary of what happened. Click any session to replay the prompt, the insight, and what it touched.
Live sessions
streamingOne graph of your whole AI-coding org
Live sessions, dead ends, root causes — the work that never reaches git becomes shared context your whole team can see, and ask.
What git never sees — who’s deep in a file right now, the upgrade that got reverted, the root cause nobody committed. Sessions remember it all.
Stop solving twice — @dmitri hits a bug, octarin recalls that @priya traced the same one two weeks ago — he skips a day of digging.
Ask octarin — “who’s hit this flake before?” answered with the person, the session, and what they found.
Your agent asks who’s been in this file
Add octarin’s MCP server to Claude Code or Cursor. Mid-task, the agent searches your team’s past sessions — who touched this, what they hit, how it ended — instead of rediscovering it.
A demonstration: inside a Claude Code session, the agent calls octarin’s MCP tools — search_sessions, get_session, graph_search — and answers questions like “have we seen this bug before?” by citing a teammate’s earlier session and its root cause.
Asks before it digs — the agent checks for prior art the way a careful engineer pings the channel: who’s touched this, what broke, whether it got fixed.
Grounded, not generated — nine tools, from search_sessions to get_session_transcript. Every answer names a teammate and cites the actual trace.
One line to connect — claude mcp add and the context flows. Works in any MCP client your team already runs.
Your agents remember. So your whole team does.
Octarin turns every session into durable memory — the decisions, conventions and gotchas worth keeping — then hands it back to any agent, on any machine, through one private MCP link. Written once, shared by default, recalled by meaning.
Octarin writes them for you
As each session is scored, Octarin distills the durable calls — “use the transaction pooler,” “de-fold cached tokens at ingest” — into structured memory. No note-taking, no prompting. Add your own, and forget what's no longer true.
Shared by default
One engineer's hard-won lesson reaches the whole team — and every future session. Org-wide by default, narrowable to a group, or kept personal. The org learns once, not once per person.
Recalled anywhere
Any MCP agent — Claude Code, Cursor, Codex, Antigravity, GitHub Copilot, Hermes, OpenClaw — finds the right memory by meaning and starts each session already caught up. One private link, every editor, every machine.
Not a file on one laptop. Memory lives in your org's cloud, not a local database — so it's there on every machine and every teammate's editor, the moment they connect.
No new infrastructure. It rides the capture you already installed. No separate database, vector index, or sync service to stand up and babysit.
Governed, not a free-for-all. Org-scoped and access-controlled, with personal scopes that stay personal and a one-click forget. SOC 2 Type I & II.
From install to answers in one afternoon
No dashboards to wire up, no SDK to embed in your code. Hook your agents once and the context builds itself.
Paste one prompt
Run a single install command. Octarin hooks Cursor, Claude Code, Codex, Antigravity, GitHub Copilot, Hermes, and OpenClaw with no per-editor setup and no source code leaving your machine.
Sessions stream in
Every agent run shows up as a trace within seconds: who ran it, which model, which repo, how many tokens, and what it cost.
Ask from anywhere
Add the octarin MCP server and any agent can ask: who's hit this bug, who knows this file, what did it cost. Answers cite real teammate sessions.
Capture everywhere, recall anywhere
Hooks for the agents your team already runs. One MCP endpoint for everything that wants the context back.
Native hooks. No SDK, no proxy, no editor lag.
Full traces over native hooks: every turn, tool call, token.
Sessions captured with real content — prompts, edits, shell.
Hooked the same way. Every run lands in the graph.
Google's agentic IDE — captured on session end via its hook.
VS Code over MCP, plus the cloud coding agent's sessions.
NousResearch's agent — hooked at session end.
Captured through an in-process internal hook.
One endpoint for everything that wants the context back.
Nine tools at api.octarin.ai/v1/mcp. Any MCP client — including the agents on the left.
Plain-language answers in the dashboard, every claim linked to a trace.
Adjacent systems, joined to the sessions behind them.
PRs and commits, linked to the sessions that produced them.
Issues, linked to the sessions that closed them.
Answers where the questions already happen.
The whole org, one source of truth
Every session searchable, every claim linked to a trace. Octarin keeps the context attached and the numbers honest.
A profile for every member
Tokens, cost, preferred model, favorite repos, and recent sessions, attributed to the person who ran them. Cursor usage is estimated from session content so nobody falls off the map.
Spend, included
Every trace carries its price. Roll it up per person, per repo, per model — the audit other tools sell, built in.
A live trace feed
Watch agent runs land in real time, the closest thing to looking over the whole team's shoulder at once.
Ask in plain language
"Who's worked on checkout.ts, and what did they hit?" Octarin answers with the person, the session, and the finding — no SQL and no spreadsheet exports.
One number, broken down three ways
And yes, the receipts are included — every dollar your team spends on AI coding, attributed to a person, a repository, and the model that ran it.
Who spent it
Cost, tokens, preferred model and recent sessions for every member.
Where it went
Repositories ranked by spend, with token and session counts beside them.
What ran it
Models compared head-to-head — price, volume, and how often they land.
Pricing that scales with your agents
A small, predictable fee on the AI spend Octarin tracks — metered on the tokens your team actually runs. Start free, upgrade when you grow. No functionality is ever locked.
Free
Try Octarin on a personal project.
- Up to 10M tokens / month
- Full dashboard, graph & memory
- 1 workspace
- Community support
Solo
For a developer who lives in their agent.
- Up to 35M tokens / month
- Everything in Free
- Unlimited history & search
- Email support
Team
Most popularFor teams running agents at scale.
- 100M tokens / month included
- then $6 per additional 1M tokens
- Unlimited members & workspaces
- Groups, Pulse & shared Memory
- Priority support
Enterprise
Unlimited scale, custom terms & control.
- Unlimited tokens
- Custom pricing & invoicing
- SSO / SAML & SOC 2 report
- Dedicated support & SLAs
Token usage = input + output tokens (cache reads excluded), summed per calendar month. Plans apply to your whole organization. Prices in USD.
Built on our own research
Octarin is built by NACE AI, a research lab training specialized models that run organizations. The work behind the product — meta-learning, memory, and knowledge graphs — is published here.
Nace Meta Agent
NEMA, our meta-agent, built a specialized AI that matched leading generalist models on the CPA exam.
Read →Enterprise Policy Injection with Metamodels
Injecting an organization's own policy directly into models with metamodels.
Read →Meta-Learning through Hypernetworks
Hypernetworks that generate task-specific parameters for efficient LLM fine-tuning.
Read →Security isn’t a feature. It’s the foundation.
Octarin is built to enterprise standards and SOC 2 Type I and Type II certified — independently audited and encrypted end to end.
AICPA · SOC 2
SOC 2 Type I
An independent auditor verified our security controls are designed correctly — access, encryption, change management.
Certified
AICPA · SOC 2
SOC 2 Type II
The same controls, tested in operation over a months-long audit window — proof they hold in practice, not just on paper.
Certified
Encrypted everywhere
TLS in transit, AES-256 at rest, secrets in a dedicated vault. Nothing travels or sits in the clear.
Read-only by design
Agents consume context through a read-only interface — they can recall what your team knows, never change it.
Never used for training
Your data belongs to you. It is never used to train models — ours or anyone else's.
Access through SSO
Sign-in through your Google Workspace identity, with scoped keys you can revoke at any time.
Setup for Agent
Point your coding agent at this
No account, no browser. The agent provisions its own workspace and starts streaming on this machine — claim it to your account anytime.
Run on the agent’s machine
npx octarin-cli@latest init --agentThen attach it to your account: octarin claim