01 every morning
A summary with no exports
«Every Monday at 9:00 send how many leads arrived, from where and for how much.» The agent visits the CRM and the ad account itself and sends it to Telegram. That used to cost an analyst an hour.
v0.9.0 · closed beta
Ask in plain words.
The agent assembles the answer
«How many leads came from ads this week and how many turned into deals?» The agent goes into the ad account, the CRM and the telephony itself, reconciles them and answers in a minute. No exports, no three browser tabs, no pivot table.
Runs on your machine or your server. The data stays with you: the only outbound traffic goes to the services you connected yourself.
- 11 Russian business systems connected
- Runs as a service on a production server
- First answer 15 minutes after install
11systems: CRM, ads, telephony, mail, finance
15minutes from install to the first answer
910tests, each about a way of losing data
0data leaves: there is no telemetry in the code
The numbers were recounted from the code on 22.08.2026, not copied from the previous version of this page.
01 / The work
What it takes off your plate
in the first week
Not «platform capabilities», but the specific jobs somebody does by hand today, usually in the evening.
02 in a minute
An answer to «where did this client come from»
Ask about one specific deal and the agent pulls the call, the form and the ad source and shows the chain. Not a hypothesis, but what is actually in the systems.
03 before the report
A check that the number is not lying
It compares how many records the service declared against how many it returned, and names the gap out loud. An incomplete export never reaches the report where nobody would catch it.
04 instead of a script
Recurring work on a schedule
A report, a reconciliation, a reminder. Set with an ordinary sentence rather than a scheduler entry, and it lives without a developer.
05 nothing to learn
It works where you already sit
Telegram, Discord, the Bitrix chat, a page in the browser. Nobody has to learn one more interface: they just ask.
06 calmly
It does nothing silently
Sending an email, editing a deal, writing off money - it shows exactly what it is about to do and waits for your «yes». Consent is single-use.
02 / The difference
A normal integration knows the API methods.
This one knows where they lie
This is the line between «the service is connected» and «the number can be trusted». Every connector is written around specific ways of losing data silently.
01
A batch request loses pages silently
Bitrix serves batch with halt=0 and keeps going after an error inside the batch. Code that only looks at the response status gets an incomplete export and treats it as complete.
batch completeness check
02
An error with status 200
Direct queues heavy reports and answers 201 or 202. Treat «not 200» as a failure and you declare keyword reports unavailable. Treat 200 as success and you never read the error in the body.
parse the body, not the status
03
Completeness that is not there
The service declared 4,812 records and returned 3,900. The honest answer is to name the gap and stop. An ordinary library returns whatever it managed to collect, and you find the difference a month later in a report.
declared against received
04
Account structure cannot be guessed
Fields, pipelines, dictionaries and smart processes differ for everyone. The agent first discovers how your portal is built and only then counts. A schema assumed in advance breaks on the second client.
discover before analysis
03 / Systems
It connects to what
you already run
Nothing has to move: the agent comes to your systems rather than inviting you into its own. Next to each one it says whether we verified it on a production account or so far only against the service's documentation.
Bitrix24
portal structure, fields, dictionaries, deals and smart processes
verified in production
Yandex.Direct
campaigns, groups, reports, spend and API limits
verified in production
UIS / Comagic
calls, forms and the quality of source tagging
verified in production
Calltouch
calls, requests and advertising sources
verified live
Web search
Yandex Search API, SearXNG and Brave under one result shape
verified live
Browser
reading script-rendered pages, and actions behind confirmation
verified live
Yandex.Mail
folders, search and reading messages
ready, awaiting your access
ZenMoney
accounts, balances and transactions; writes require confirmation
ready, awaiting your access
VK
community messages and Long Poll state
ready, awaiting your access
Roistat
analytics and comparison of attribution models
ready, awaiting your access
Alice
a voice skill as one more surface for the conversation
ready, awaiting your access
MeetScribe
meeting transcripts and minutes
ready, awaiting your access
04 / How it runs
From question to answer
01
A question in plain words
Telegram, Discord, the browser panel, a web page or the terminal. No command to learn.
02
The agent picks a source
It looks at which services are connected and decides where to go. What is not on the list it does not think about and does not invent.
03
Structure discovery
First it works out how your account is built: fields, pipelines, dictionaries. Only then does it count.
04
Completeness check
It compares the actual output against the declared one, catches errors in the response body, accounts for limits and pagination. A gap is named, not smoothed over.
05
Answer and trail
It shows which systems it visited and what the answer rests on. Recurring work goes on a schedule, important things are remembered for the long run.
05 / Trust
Why this can be let
near a production CRM
The audit on 19.08.2026 closed 21 findings. Below are not promises from the model but properties of the code: each one has a test.
01 perimeter
The data stays with you
The agent runs on your machine or your server. History, notes and jobs sit next to it as ordinary files. Telemetry sending does not exist in the code.
02 confirmations
Dangerous work is held
Anything that changes data in an external service is shown before it runs: the exact operation name and its arguments. Consent is single-use and only inside its own chat.
03 injection
Someone else's data is data, not commands
The contents of emails and records are wrapped in a boundary with a random marker. An instruction that arrives inside a stranger's email does not become a command.
04 isolation
A connector sees only its own
Each connector receives only its own environment variables. Credentials for one service are invisible to the connector of another.
05 shell
Shell commands are not executed
A deliberate decision, not an unfinished feature. An agent with access to other people's systems should not be able to run anything it likes on the machine.
06 fatigue
It asks where a mistake has a price
Reading, searching and parsing need no confirmation. A question asked for the twentieth time in a day turns into a «yes» button that gets pressed without looking.
06 / Getting started
Fifteen minutes
to the first answer
One Go binary and connectors on the Python standard library. No Docker, no cluster, no runtime. Step by step on the quick start page.
bash
# build and configure$ git clone https://github.com/nobi-k/interclaw && cd interclaw$ make build && ./bin/interclaw setup # ask in plain words$ ./bin/interclaw serve> how many leads arrived this week and from which sources?138 enquiries. Ads 41, marketplace 36, direct 44, organic 17.the portal declared 142 and returned 138: four records have no source, show them?
That last line is the difference. An ordinary export would have returned 138 and said nothing.
07 / Honestly
What already works,
and what does not yet
This list stays on the public page rather than in private correspondence. Finding out about an unfinished part in week three of a pilot suits nobody, and us least of all.
Linux and server
the main platform: service, schedule, channels, every connector
works
Web interface and extension
a chat page served by the core itself and a panel beside the site
works
Windows
the cross-build and installer are produced, we have no live machine to run them on
not verified
Android
a client skeleton in the repository; calling it an application would be early
skeleton
Cloud
the agent on our hardware: under construction, waiting list open
in development
Multi-tenancy
this version targets a personal or dedicated perimeter, not a shared SaaS
not included
08 / Questions
What people usually ask
How long will it take to launch with us?
Installing it and getting the first answer takes fifteen minutes. Getting to a daily scenario on your own data takes a two to three week pilot: the time goes into access, structure discovery, and checking that the numbers match what you see with your own eyes.
Does our data go to your server?
No. The agent runs on your machine, history and notes sit in a local directory. The only outbound traffic goes to the services you connected yourself and to the model provider you chose.
Our Bitrix is heavily customised. Will it work?
That is exactly what it was written for. The agent assumes no schema in advance: it first discovers your portal - fields, pipelines, dictionaries, smart processes - and only then counts. A schema guessed in advance breaks on the second client.
How is this different from ChatGPT with access to our data?
The data does not leave, and the connectors are written around the specific ways specific services lose data. Everybody can build a shell today; catching the silent loss of half an export is rarer.
What if the model makes a mistake and damages our data?
It cannot. Anything that changes data in an external service is held and shown to you before it runs, with the exact operation name and arguments. The agent does not execute shell commands at all.
What does it cost?
Diagnostics 10,000 RUB, pilot 50,000 RUB, then support at 35,000 RUB per month. Installing it yourself and using it is free. Details in pricing.
09 / Next
We will show it on your data
Twenty minutes of conversation: you describe which systems you run and what you count by hand, we say what an agent will close and what it will not. No deck, and no «leave a request, a manager will be in touch».