Takibi Base Shared Knowledge Base: Cited Retrieval

Takibi Base is a shared knowledge base for AI agents: a team organizes its own documents, then gives each connected tool a Profile that controls which folders or projects it can read. When an agent asks through the CLI or API, Takibi returns matching source passages, citations, and an evidence-support score. It does not write the final answer. That makes the Takibi Base shared knowledge base a retrieval layer for controlled, source-grounded workflows—not a chatbot to use on its own.

Last updated: October 9, 2026.

Quick facts

• Pricing: Paid plans from €18/month plus VAT; seven-day, card-required trial.
• Platform: Web app, CLI, and API; CLI requires Node.js 20+.
• Best for: Teams giving multiple AI agents scoped access to shared documents.
• Free tier: No permanent free plan; trial only.
• Verified: October 2026.

How the Takibi Base shared knowledge base works

Takibi puts a human-managed document library between source material and AI agents. Its emphasis is permissioned retrieval and inspectable evidence, not a chat screen that writes answers. Four parts of that workflow matter most.

1. Organize files and keep the originals

People sort documents into workspaces, projects, and folders. The regular uploader accepts text-based PDFs, Word .docx, Markdown, and plain text up to 10 MB per file. Takibi indexes supported text, retains original files for download, and can export a project in Open Knowledge Format. Scanned PDFs need separate OCR, while legacy .doc files are unsupported. Document status distinguishes indexed, converting, and quarantined files.

2. Scope agent access with Profiles

A Profile holds an agent's document grants, keys, capabilities, and usage budgets. Folder grants cover indexed files in the selected folder; project grants also cover future folders and unfiled material. Every key attached to one Profile shares its permissions. Teams can create separate Profiles for agents that need different access, rotate or revoke individual keys, pause all keys on a Profile, and review activity.

That boundary needs deliberate setup: a second key is not a separate permission set if it belongs to the same Profile. Start with the narrowest folder grant that still supports the agent's task, and use a project grant only when the wider scope is intended.

3. Return evidence, not answer prose

Through the CLI or API, an agent can ask a question and receive matching source passages, citations, and an evidence-support score. If Takibi finds no matching evidence, it returns no passages. The connected agent—not Takibi—must compose the final response.

The score is a retrieval signal, not a probability that a claim is true. The vendor also cautions that citations and automated judgments can be incomplete or wrong, so readers should check the cited source. See the official agent connection documentation for the CLI/API setup.

4. Keep notes separate from approved sources

Optional Notes belong to a Profile, so its keys can share working observations while other Profiles cannot. Each note expires 30 days after creation and never becomes a source document automatically. A person can review its claims and citations before adding material to the library. Project tasks also let agents claim work, report progress, and submit results for review.

That separation is useful when an agent records a temporary discovery: it remains a lead to check, not a new policy silently added to the team's knowledge base.

Takibi Base pricing (verified October 2026)

Takibi lists two monthly plans for one human operator and one workspace. Ember is €18/month for two Profiles, 250 asks per day, and 300,000 tokens per day. Campfire is €52/month for eight Profiles, 800 asks per day, and 800,000 tokens per day. Both plans include the same product tools; daily allowances are shared at workspace level, with Profile budgets also available. Prices exclude VAT.

There is no ongoing free plan. Each tier begins with a seven-day trial that requires a card; billing starts after the trial unless canceled. Additional human seats are not included in the current version. Check the official Takibi pricing page for the checkout tax and renewal terms.

Fit, limitations, and data handling

Takibi fits teams whose agent can run a CLI or make authenticated HTTP requests. The CLI needs Node.js 20+, a scoped key, and a place in the agent's environment to store it. A normal chat interface without tool access cannot connect unaided. That technical setup is the main adoption hurdle.

Review data handling before importing sensitive records. The published policy names a hosted provider in Frankfurt and Western Europe, plus TypeSafe/Jev for some AI judgments. Depending on the feature, a request may include questions, document excerpts or text, filenames, and context; that AI processor is not described as EU-only. Ordinary document listing and text search do not call Jev. Cancellation does not erase workspace data, and the policy sets no fixed deletion deadline across every provider. The service is run by a named sole proprietor in Spain; a data-processing agreement can be requested. For regulated information, obtain the assurances your organization needs before upload.

The source-based fit is strongest when citations and per-agent boundaries matter more than an all-in-one chat experience. Skip this if you need Takibi itself to draft answers, cannot maintain a CLI/API integration, or cannot send queries or excerpts to an external AI processor.

Pros and cons

Alternatives: when to choose another tool

• Gemini Notebook: Pick it for a ready-made research notebook that answers questions from provided sources. Takibi is the better fit when multiple connected agents need Profile-level document permissions and passage retrieval.
• AnythingLLM: Pick it when a local document assistant that generates chat responses matters more than a neutral retrieval layer. Its current product positioning emphasizes on-device use and self-hosted multi-user deployment; Takibi separates retrieval from the agent that writes the answer.

For adjacent listings, browse the Research and Productivity categories. If you are choosing the general assistant that might sit downstream, compare our Claude Sonnet 5.5 and ChatGPT listings; these are separate products, not built-in Takibi connectors.

Frequently asked questions

What is Takibi Base used for?

Takibi Base stores a team's documents in projects and folders, then lets connected AI agents retrieve passages from sources allowed by a Profile. Its CLI or API returns citations and an evidence-support score. It is a retrieval layer; the connected agent, not Takibi, writes any final answer.

Does Takibi Base generate AI answers?

No. Takibi returns source passages, citations, and evidence signals, then leaves response writing to the connected agent. When it finds no matching evidence, it can return no passages. The support score measures evidence support; the documentation says it is not a probability that a statement is true.

Does Takibi Base have a free plan?

No permanent free plan is listed. Ember starts at €18 monthly and Campfire at €52 monthly, before VAT. Both begin with a seven-day trial that requires a payment card; the subscription starts charging after the trial unless the customer cancels in time through billing settings.

Which file formats does Takibi Base accept?

The regular uploader supports text-based PDFs, DOCX, Markdown, and plain text up to 10 MB per file. Scanned PDFs need OCR before upload, and legacy DOC files are not supported. Import with AI handles Markdown or text collections, so use the regular uploader for PDF and DOCX.

Are Takibi Base documents private?

Takibi uses hosted infrastructure and an AI-processing subprocessor for some retrieval, import, and content-check features; the provider may receive questions, excerpts, or filenames and is not described as EU-only. Review the privacy terms and request a DPA before uploading sensitive company records. Cancellation also does not erase workspace data automatically.