Echo is an open-source (AGPLv3) full-cycle discovery agent for the people enterprise tools price out — independent researchers, small NGOs, solo operators, artist-technologists, and small founders entering unfamiliar domains. It runs one loop end to end: discover → classify → synthesise → draft outward signals → post them to real channels → listen for what returns → accumulate what it learns over time. Continuous discovery runs at zero marginal cost on rotated free-tier models; frontier-grade synthesis happens only in an explicit, operator-triggered boost inside a subscription you already pay for.
Echo is the open-source generalisation of Khashif — a working, two-month-proven private discovery agent — released so anyone can build their own for their own domain.
| What | A full-cycle discovery & intelligence toolkit (discover → act → listen → accumulate) |
| For | Solo operators, independent researchers, small NGOs, artist-technologists, small founders |
| Mechanism | Free-tier cost-cascade for continuous work + operator-triggered frontier boost for synthesis |
| Cost | Zero new recurring AI spend — boost runs inside a subscription you already maintain |
| License | AGPLv3 + optional dual licensing for commercial integrations |
| Status | Skeleton — v1 is the subject of an NLnet NGI Zero Commons Fund application (June 2026); public code lands during the funded development cycle |
| Sibling | Cairn — coordination integrity for multi-agent AI systems (same operator, same commons philosophy) |
- Free-tier mode — the always-on default: providers (Groq, Cerebras, Gemini) rotated by role do continuous discovery, classification, multi-hop crawl, and link accumulation at zero marginal cost, with no frontier model involved.
- Boost mode — an explicit, operator-triggered single cycle that runs inside the operator's own frontier subscription (Claude Code, ChatGPT, Cursor, or a compatible local model) for synthesis moments that need frontier reach.
- Give-and-take (vermek-almak) — the relationship with the frontier: Echo gives it persistent memory, accumulated field topology, and a return-listening loop across time and channels; the frontier gives Echo synthesis-grade reasoning at the moments that matter.
- Intersection intelligence (water + flour = bread) — a primitive for surfacing combinations of two domains that produce an opportunity neither could alone, filtered by four hard rules (novelty, mechanism, buyer, why-now) so shallow pairings die.
- Echo layer — the listening half: every outbound signal becomes a watched thread, and the layer folds the return (a comment, a reply, a state change) back into the operator's view, so outreach stops disappearing into silence.
Echo runs in two cleanly separated modes — never together, never blended in a single decision (biri ya da öbürü): free-tier mode for continuous, zero-cost work, and boost mode for frontier synthesis on demand. The two are separated at runtime, linked at memory — each focuses on what the other cannot do, and both feed the same accumulated pool over time.
- Free-tier mode is the substrate's solar panel — periodic free flow over weeks and months, quietly accumulating links, classifications, returns, and feedback the frontier could never harvest in one session.
- Boost mode is when that accumulated context concentrates into an act of synthesis — when called, the frontier draws from the pool the substrate has been building all along.
The dial is in the operator's hand, not the vendor's: pay only when paying matters, and only for intelligence you already pay for. No new paid API integration is stacked on your existing subscriptions — distinct from per-API-call SaaS agents (Lindy, LangChain Cloud, Manus) that wrap a frontier model behind their own credit meter.
These crystallised through the originating system's real two-month use:
- Multi-LLM cost cascade with operator-controlled boost — Echo as the extended hand of the frontier (the two-mode design above).
- Context-aware multi-hop crawling that learns from past resonance — feeds and discovery queries seed a bounded crawl that follows resonant links within the operator's domain; past high-scoring items shape future seed queries.
- Intersection-intelligence formulation that accumulates over time — the water + flour = bread primitive; each surviving intersection is added to a persistent label memory so the system never proposes the same pairing twice.
- Full-cycle outbound + echo layer — Echo doesn't merely discover and report, it acts: reference channels ship for GitHub issues, email (Resend), and a task manager (Remember The Milk); every outbound signal becomes a watched thread whose return is folded back in.
- Cumulative learning across journeys — keywords, intersections, echoes, operator decisions, visited feeds, crawled pages persist into a structured memory that grows with use; each new journey starts from accumulated context, not from zero.
- Phase 0 / Phase 2 — discovery search + bounded multi-hop crawl engine
- Phase 3 — intersection-intelligence formulation
- Phase 6 — echo layer (
watched_threads,WATCH_HANDLERSdispatch) - Multi-LLM cost-cascade interface (free-tier rotation + paid boost on demand)
- Pluggable storage: sqlite (default) + Supabase (reference adapter)
- Pluggable outbound: GitHub issues + Resend + RTM (reference adapters)
- YAML configuration schema
- Reference demo application: academic-research radar (arXiv papers + open grant calls + author/lab intersections)
- Documentation, manifesto, examples
Out of v1 (deferred to v2): Phase 4 deep-dive briefings, Phase 5 broadcast / giver layer, additional storage adapters (Postgres native, Redis), additional outbound channels, web dashboard.
What is Echo, in one sentence? An open-source (AGPLv3) full-cycle discovery agent that discovers, classifies, synthesises, acts, and listens for returns — built for solo operators and small teams, the open generalisation of the private Khashif agent.
Is it expensive to run? No. Continuous work runs at zero marginal cost on rotated free-tier models; frontier synthesis happens only in an explicit boost inside a subscription you already pay for. No new recurring AI integration.
How is it different from Lindy / LangChain Cloud / Manus? Those wrap a frontier model behind their own per-call credit meter. Echo separates a free-tier substrate (always-on, accumulating) from operator-triggered frontier boost — the dial is in your hand, and you pay only when synthesis matters.
Can I run it now? Echo is at skeleton status. Full v1 development is the subject of an NLnet NGI Zero Commons Fund application submitted June 2026; public code lands during the funded development cycle (or via the September 2026 retry cycle / an alternative grant track).
What's the license? AGPLv3 — strong copyleft protects derivative work from re-closure, and the network-use clause ensures hosted services contribute modifications back. Dual licensing is available for commercial integrations that cannot accept AGPL.
How does it relate to Cairn? Cairn (github.com/echo-toolkit/cairn) is the sibling project — coordination integrity for multi-agent AI systems. Same operator, same commons philosophy, independent projects.
AGPLv3. Strong copyleft protects derivative work from re-closure; the network-use clause ensures hosted services contribute modifications back. Dual licensing is available for commercial integrations that cannot accept AGPL's terms — contact below.
- Khashif — the private originating system. Two months of intense iteration produced the six-phase architecture Echo generalises. Khashif itself remains private — the operator's accumulated network, intersections, and proprietary configuration. Echo is the infrastructure; Khashif is one operator's configuration of it.
- Cairn — sibling project under the same org: coordination integrity for multi-agent AI systems (passive append-only blackboard + minimal-context workers). Same operator, same commons philosophy, independent projects.
- Casa Caravan — the operator's artist surface: sound work, music, gong-making, audio sanctuary.
Tağmaç Çankaya — Lefkoşa, Cyprus (Cyprus Republic / EU citizen). Contact: tagmacc@gmail.com. GitHub Sponsors / commercial relicensing inquiries: same address.
Echo — an open-source (AGPLv3) full-cycle discovery & intelligence toolkit for solo operators and small teams: discover → classify → synthesise → act → listen → accumulate, with a free-tier cost cascade and operator-triggered frontier boost. The open generalisation of the Khashif discovery agent. https://github.com/echo-toolkit/echo
- Code: https://github.com/echo-toolkit/echo
- Originating system: Khashif — https://khashif.run
- Sibling project: Cairn — https://github.com/echo-toolkit/cairn
- Author: Tağmaç Çankaya — Lefkoşa, Cyprus / EU
Echo · MMXXVI · the call and the echo